diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file
diff --git a/ai-hub/app/core/services/tts.py b/ai-hub/app/core/services/tts.py
new file mode 100644
index 0000000..c196795
--- /dev/null
+++ b/ai-hub/app/core/services/tts.py
@@ -0,0 +1,25 @@
+from typing import AsyncGenerator
+from app.core.providers.base import TTSProvider
+
+class TTSService:
+    """
+    Service class for generating speech from text using a TTS provider.
+    """
+    def __init__(self, tts_provider: TTSProvider):
+        """
+        Initializes the TTSService with a concrete TTS provider.
+        """
+        self.tts_provider = tts_provider
+
+    async def create_speech_stream(self, text: str) -> AsyncGenerator[bytes, None]:
+        """
+        Generates a stream of audio bytes from the given text using the configured
+        TTS provider.
+
+        Args:
+            text: The text to be converted to speech.
+
+        Returns:
+            An async generator that yields chunks of audio bytes.
+        """
+        return self.tts_provider.generate_speech(text)
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file
diff --git a/ai-hub/app/core/services/tts.py b/ai-hub/app/core/services/tts.py
new file mode 100644
index 0000000..c196795
--- /dev/null
+++ b/ai-hub/app/core/services/tts.py
@@ -0,0 +1,25 @@
+from typing import AsyncGenerator
+from app.core.providers.base import TTSProvider
+
+class TTSService:
+    """
+    Service class for generating speech from text using a TTS provider.
+    """
+    def __init__(self, tts_provider: TTSProvider):
+        """
+        Initializes the TTSService with a concrete TTS provider.
+        """
+        self.tts_provider = tts_provider
+
+    async def create_speech_stream(self, text: str) -> AsyncGenerator[bytes, None]:
+        """
+        Generates a stream of audio bytes from the given text using the configured
+        TTS provider.
+
+        Args:
+            text: The text to be converted to speech.
+
+        Returns:
+            An async generator that yields chunks of audio bytes.
+        """
+        return self.tts_provider.generate_speech(text)
\ No newline at end of file
diff --git a/ai-hub/integration_tests/conftest.py b/ai-hub/integration_tests/conftest.py
new file mode 100644
index 0000000..5d51010
--- /dev/null
+++ b/ai-hub/integration_tests/conftest.py
@@ -0,0 +1,57 @@
+import httpx
+import pytest_asyncio
+
+BASE_URL = "http://127.0.0.1:8000"
+
+@pytest_asyncio.fixture(scope="session")
+def base_url():
+    """Fixture to provide the base URL for the tests."""
+    return BASE_URL
+
+@pytest_asyncio.fixture(scope="function") # <-- Change scope to "function"
+async def http_client():
+    """
+    Fixture to provide an async HTTP client for all tests in the session.
+    A new client is created and closed properly using a try/finally block
+    to prevent "Event loop is closed" errors.
+    """
+    client = httpx.AsyncClient(base_url=BASE_URL, timeout=60.0)
+    try:
+        yield client
+    finally:
+        await client.aclose()
+
+@pytest_asyncio.fixture(scope="function")
+async def session_id(http_client):
+    """
+    Creates a new session before a test and cleans it up after.
+    Returns the session ID.
+    """
+    payload = {"user_id": "integration_tester", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    yield session_id
+    # No explicit session deletion is needed for this example,
+    # as sessions are typically managed by a database lifecycle.
+
+@pytest_asyncio.fixture(scope="function")
+async def document_id(http_client):
+    """
+    Creates a new document before a test and ensures it's deleted afterward.
+    Returns the document ID.
+    """
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    response = await http_client.post("/documents", json=doc_data)
+    assert response.status_code == 200
+    try:
+        message = response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+        
+    yield document_id
+    
+    # Teardown: Delete the document after the test
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file
diff --git a/ai-hub/app/core/services/tts.py b/ai-hub/app/core/services/tts.py
new file mode 100644
index 0000000..c196795
--- /dev/null
+++ b/ai-hub/app/core/services/tts.py
@@ -0,0 +1,25 @@
+from typing import AsyncGenerator
+from app.core.providers.base import TTSProvider
+
+class TTSService:
+    """
+    Service class for generating speech from text using a TTS provider.
+    """
+    def __init__(self, tts_provider: TTSProvider):
+        """
+        Initializes the TTSService with a concrete TTS provider.
+        """
+        self.tts_provider = tts_provider
+
+    async def create_speech_stream(self, text: str) -> AsyncGenerator[bytes, None]:
+        """
+        Generates a stream of audio bytes from the given text using the configured
+        TTS provider.
+
+        Args:
+            text: The text to be converted to speech.
+
+        Returns:
+            An async generator that yields chunks of audio bytes.
+        """
+        return self.tts_provider.generate_speech(text)
\ No newline at end of file
diff --git a/ai-hub/integration_tests/conftest.py b/ai-hub/integration_tests/conftest.py
new file mode 100644
index 0000000..5d51010
--- /dev/null
+++ b/ai-hub/integration_tests/conftest.py
@@ -0,0 +1,57 @@
+import httpx
+import pytest_asyncio
+
+BASE_URL = "http://127.0.0.1:8000"
+
+@pytest_asyncio.fixture(scope="session")
+def base_url():
+    """Fixture to provide the base URL for the tests."""
+    return BASE_URL
+
+@pytest_asyncio.fixture(scope="function") # <-- Change scope to "function"
+async def http_client():
+    """
+    Fixture to provide an async HTTP client for all tests in the session.
+    A new client is created and closed properly using a try/finally block
+    to prevent "Event loop is closed" errors.
+    """
+    client = httpx.AsyncClient(base_url=BASE_URL, timeout=60.0)
+    try:
+        yield client
+    finally:
+        await client.aclose()
+
+@pytest_asyncio.fixture(scope="function")
+async def session_id(http_client):
+    """
+    Creates a new session before a test and cleans it up after.
+    Returns the session ID.
+    """
+    payload = {"user_id": "integration_tester", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    yield session_id
+    # No explicit session deletion is needed for this example,
+    # as sessions are typically managed by a database lifecycle.
+
+@pytest_asyncio.fixture(scope="function")
+async def document_id(http_client):
+    """
+    Creates a new document before a test and ensures it's deleted afterward.
+    Returns the document ID.
+    """
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    response = await http_client.post("/documents", json=doc_data)
+    assert response.status_code == 200
+    try:
+        message = response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+        
+    yield document_id
+    
+    # Teardown: Delete the document after the test
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_documents.py b/ai-hub/integration_tests/test_documents.py
new file mode 100644
index 0000000..c6d4362
--- /dev/null
+++ b/ai-hub/integration_tests/test_documents.py
@@ -0,0 +1,33 @@
+import pytest
+
+@pytest.mark.asyncio
+async def test_document_lifecycle(http_client):
+    """
+    Tests the full lifecycle of a document: add, list, and delete.
+    This is run as a single, sequential test for a clean state.
+    """
+    print("\n--- Running test_document_lifecycle ---")
+    
+    # 1. Add a new document
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    add_response = await http_client.post("/documents", json=doc_data)
+    assert add_response.status_code == 200
+    try:
+        message = add_response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    print(f"✅ Document for lifecycle test created with ID: {document_id}")
+    
+    # 2. List all documents and check if the new document is present
+    list_response = await http_client.get("/documents")
+    assert list_response.status_code == 200
+    ids_in_response = {doc["id"] for doc in list_response.json()["documents"]}
+    assert document_id in ids_in_response
+    print("✅ Document list test passed.")
+    
+    # 3. Delete the document
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
+    assert delete_response.json()["document_id"] == document_id
+    print("✅ Document delete test passed.")
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file
diff --git a/ai-hub/app/core/services/tts.py b/ai-hub/app/core/services/tts.py
new file mode 100644
index 0000000..c196795
--- /dev/null
+++ b/ai-hub/app/core/services/tts.py
@@ -0,0 +1,25 @@
+from typing import AsyncGenerator
+from app.core.providers.base import TTSProvider
+
+class TTSService:
+    """
+    Service class for generating speech from text using a TTS provider.
+    """
+    def __init__(self, tts_provider: TTSProvider):
+        """
+        Initializes the TTSService with a concrete TTS provider.
+        """
+        self.tts_provider = tts_provider
+
+    async def create_speech_stream(self, text: str) -> AsyncGenerator[bytes, None]:
+        """
+        Generates a stream of audio bytes from the given text using the configured
+        TTS provider.
+
+        Args:
+            text: The text to be converted to speech.
+
+        Returns:
+            An async generator that yields chunks of audio bytes.
+        """
+        return self.tts_provider.generate_speech(text)
\ No newline at end of file
diff --git a/ai-hub/integration_tests/conftest.py b/ai-hub/integration_tests/conftest.py
new file mode 100644
index 0000000..5d51010
--- /dev/null
+++ b/ai-hub/integration_tests/conftest.py
@@ -0,0 +1,57 @@
+import httpx
+import pytest_asyncio
+
+BASE_URL = "http://127.0.0.1:8000"
+
+@pytest_asyncio.fixture(scope="session")
+def base_url():
+    """Fixture to provide the base URL for the tests."""
+    return BASE_URL
+
+@pytest_asyncio.fixture(scope="function") # <-- Change scope to "function"
+async def http_client():
+    """
+    Fixture to provide an async HTTP client for all tests in the session.
+    A new client is created and closed properly using a try/finally block
+    to prevent "Event loop is closed" errors.
+    """
+    client = httpx.AsyncClient(base_url=BASE_URL, timeout=60.0)
+    try:
+        yield client
+    finally:
+        await client.aclose()
+
+@pytest_asyncio.fixture(scope="function")
+async def session_id(http_client):
+    """
+    Creates a new session before a test and cleans it up after.
+    Returns the session ID.
+    """
+    payload = {"user_id": "integration_tester", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    yield session_id
+    # No explicit session deletion is needed for this example,
+    # as sessions are typically managed by a database lifecycle.
+
+@pytest_asyncio.fixture(scope="function")
+async def document_id(http_client):
+    """
+    Creates a new document before a test and ensures it's deleted afterward.
+    Returns the document ID.
+    """
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    response = await http_client.post("/documents", json=doc_data)
+    assert response.status_code == 200
+    try:
+        message = response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+        
+    yield document_id
+    
+    # Teardown: Delete the document after the test
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_documents.py b/ai-hub/integration_tests/test_documents.py
new file mode 100644
index 0000000..c6d4362
--- /dev/null
+++ b/ai-hub/integration_tests/test_documents.py
@@ -0,0 +1,33 @@
+import pytest
+
+@pytest.mark.asyncio
+async def test_document_lifecycle(http_client):
+    """
+    Tests the full lifecycle of a document: add, list, and delete.
+    This is run as a single, sequential test for a clean state.
+    """
+    print("\n--- Running test_document_lifecycle ---")
+    
+    # 1. Add a new document
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    add_response = await http_client.post("/documents", json=doc_data)
+    assert add_response.status_code == 200
+    try:
+        message = add_response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    print(f"✅ Document for lifecycle test created with ID: {document_id}")
+    
+    # 2. List all documents and check if the new document is present
+    list_response = await http_client.get("/documents")
+    assert list_response.status_code == 200
+    ids_in_response = {doc["id"] for doc in list_response.json()["documents"]}
+    assert document_id in ids_in_response
+    print("✅ Document list test passed.")
+    
+    # 3. Delete the document
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
+    assert delete_response.json()["document_id"] == document_id
+    print("✅ Document delete test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_misc.py b/ai-hub/integration_tests/test_misc.py
new file mode 100644
index 0000000..3ff7925
--- /dev/null
+++ b/ai-hub/integration_tests/test_misc.py
@@ -0,0 +1,36 @@
+import pytest
+import httpx
+
+@pytest.mark.asyncio
+async def test_root_endpoint(http_client):
+    """
+    Tests if the root endpoint is alive and returns the correct status message.
+    """
+    print("\n--- Running test_root_endpoint ---")
+    response = await http_client.get("/")
+    assert response.status_code == 200
+    assert response.json() == {"status": "AI Model Hub is running!"}
+    print("✅ Root endpoint test passed.")
+
+# @pytest.mark.asyncio
+# async def test_create_speech_stream(http_client):
+#     """
+#     Tests the /speech endpoint for a successful audio stream response.
+#     """
+#     print("\n--- Running test_create_speech_stream ---")
+#     url = "/speech"
+#     payload = {"text": "Hello, world!"}
+
+#     # The `stream=True` parameter tells httpx to not read the entire response body
+#     # at once. We'll handle it manually to check for content.
+#     async with http_client.stream("POST", url, json=payload) as response:
+#         assert response.status_code == 200, f"Speech stream request failed. Response: {response.text}"
+#         assert response.headers.get("content-type") == "audio/wav"
+
+#         # Check that the response body is not empty by iterating over chunks.
+#         content_length = 0
+#         async for chunk in response.aiter_bytes():
+#             content_length += len(chunk)
+            
+#         assert content_length > 0
+#     print("✅ TTS stream test passed.")
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file
diff --git a/ai-hub/app/core/services/tts.py b/ai-hub/app/core/services/tts.py
new file mode 100644
index 0000000..c196795
--- /dev/null
+++ b/ai-hub/app/core/services/tts.py
@@ -0,0 +1,25 @@
+from typing import AsyncGenerator
+from app.core.providers.base import TTSProvider
+
+class TTSService:
+    """
+    Service class for generating speech from text using a TTS provider.
+    """
+    def __init__(self, tts_provider: TTSProvider):
+        """
+        Initializes the TTSService with a concrete TTS provider.
+        """
+        self.tts_provider = tts_provider
+
+    async def create_speech_stream(self, text: str) -> AsyncGenerator[bytes, None]:
+        """
+        Generates a stream of audio bytes from the given text using the configured
+        TTS provider.
+
+        Args:
+            text: The text to be converted to speech.
+
+        Returns:
+            An async generator that yields chunks of audio bytes.
+        """
+        return self.tts_provider.generate_speech(text)
\ No newline at end of file
diff --git a/ai-hub/integration_tests/conftest.py b/ai-hub/integration_tests/conftest.py
new file mode 100644
index 0000000..5d51010
--- /dev/null
+++ b/ai-hub/integration_tests/conftest.py
@@ -0,0 +1,57 @@
+import httpx
+import pytest_asyncio
+
+BASE_URL = "http://127.0.0.1:8000"
+
+@pytest_asyncio.fixture(scope="session")
+def base_url():
+    """Fixture to provide the base URL for the tests."""
+    return BASE_URL
+
+@pytest_asyncio.fixture(scope="function") # <-- Change scope to "function"
+async def http_client():
+    """
+    Fixture to provide an async HTTP client for all tests in the session.
+    A new client is created and closed properly using a try/finally block
+    to prevent "Event loop is closed" errors.
+    """
+    client = httpx.AsyncClient(base_url=BASE_URL, timeout=60.0)
+    try:
+        yield client
+    finally:
+        await client.aclose()
+
+@pytest_asyncio.fixture(scope="function")
+async def session_id(http_client):
+    """
+    Creates a new session before a test and cleans it up after.
+    Returns the session ID.
+    """
+    payload = {"user_id": "integration_tester", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    yield session_id
+    # No explicit session deletion is needed for this example,
+    # as sessions are typically managed by a database lifecycle.
+
+@pytest_asyncio.fixture(scope="function")
+async def document_id(http_client):
+    """
+    Creates a new document before a test and ensures it's deleted afterward.
+    Returns the document ID.
+    """
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    response = await http_client.post("/documents", json=doc_data)
+    assert response.status_code == 200
+    try:
+        message = response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+        
+    yield document_id
+    
+    # Teardown: Delete the document after the test
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_documents.py b/ai-hub/integration_tests/test_documents.py
new file mode 100644
index 0000000..c6d4362
--- /dev/null
+++ b/ai-hub/integration_tests/test_documents.py
@@ -0,0 +1,33 @@
+import pytest
+
+@pytest.mark.asyncio
+async def test_document_lifecycle(http_client):
+    """
+    Tests the full lifecycle of a document: add, list, and delete.
+    This is run as a single, sequential test for a clean state.
+    """
+    print("\n--- Running test_document_lifecycle ---")
+    
+    # 1. Add a new document
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    add_response = await http_client.post("/documents", json=doc_data)
+    assert add_response.status_code == 200
+    try:
+        message = add_response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    print(f"✅ Document for lifecycle test created with ID: {document_id}")
+    
+    # 2. List all documents and check if the new document is present
+    list_response = await http_client.get("/documents")
+    assert list_response.status_code == 200
+    ids_in_response = {doc["id"] for doc in list_response.json()["documents"]}
+    assert document_id in ids_in_response
+    print("✅ Document list test passed.")
+    
+    # 3. Delete the document
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
+    assert delete_response.json()["document_id"] == document_id
+    print("✅ Document delete test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_misc.py b/ai-hub/integration_tests/test_misc.py
new file mode 100644
index 0000000..3ff7925
--- /dev/null
+++ b/ai-hub/integration_tests/test_misc.py
@@ -0,0 +1,36 @@
+import pytest
+import httpx
+
+@pytest.mark.asyncio
+async def test_root_endpoint(http_client):
+    """
+    Tests if the root endpoint is alive and returns the correct status message.
+    """
+    print("\n--- Running test_root_endpoint ---")
+    response = await http_client.get("/")
+    assert response.status_code == 200
+    assert response.json() == {"status": "AI Model Hub is running!"}
+    print("✅ Root endpoint test passed.")
+
+# @pytest.mark.asyncio
+# async def test_create_speech_stream(http_client):
+#     """
+#     Tests the /speech endpoint for a successful audio stream response.
+#     """
+#     print("\n--- Running test_create_speech_stream ---")
+#     url = "/speech"
+#     payload = {"text": "Hello, world!"}
+
+#     # The `stream=True` parameter tells httpx to not read the entire response body
+#     # at once. We'll handle it manually to check for content.
+#     async with http_client.stream("POST", url, json=payload) as response:
+#         assert response.status_code == 200, f"Speech stream request failed. Response: {response.text}"
+#         assert response.headers.get("content-type") == "audio/wav"
+
+#         # Check that the response body is not empty by iterating over chunks.
+#         content_length = 0
+#         async for chunk in response.aiter_bytes():
+#             content_length += len(chunk)
+            
+#         assert content_length > 0
+#     print("✅ TTS stream test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_sessions.py b/ai-hub/integration_tests/test_sessions.py
new file mode 100644
index 0000000..44d63d7
--- /dev/null
+++ b/ai-hub/integration_tests/test_sessions.py
@@ -0,0 +1,107 @@
+import pytest
+
+# Test prompts and data
+CONTEXT_PROMPT = "Who is the CEO of Microsoft?"
+FOLLOW_UP_PROMPT = "When was he born?"
+RAG_DOC_TITLE = "Fictional Company History"
+RAG_DOC_TEXT = "The company AlphaCorp was founded in 2021 by Jane Doe. Their primary product is a smart home device called 'Nexus'."
+RAG_PROMPT = "Who founded AlphaCorp and what is their main product?"
+
+@pytest.mark.asyncio
+async def test_chat_in_session_lifecycle(http_client):
+    """
+    Tests a full session lifecycle from creation to conversational memory.
+    This test is a single, sequential unit.
+    """
+    print("\n--- Running test_chat_in_session_lifecycle ---")
+    
+    # 1. Create a new session
+    payload = {"user_id": "integration_tester_lifecycle", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    print(f"✅ Session created successfully with ID: {session_id}")
+    
+    # 2. First chat turn to establish context
+    chat_payload_1 = {"prompt": CONTEXT_PROMPT}
+    response_1 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_1)
+    assert response_1.status_code == 200
+    assert "Satya Nadella" in response_1.json()["answer"]
+    assert response_1.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 1 (context) test passed.")
+    
+    # 3. Second chat turn (follow-up) to test conversational memory
+    chat_payload_2 = {"prompt": FOLLOW_UP_PROMPT}
+    response_2 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_2)
+    assert response_2.status_code == 200
+    assert "1967" in response_2.json()["answer"]
+    assert response_2.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 2 (follow-up) test passed.")
+    
+    # 4. Cleanup (optional, but good practice if not using a test database that resets)
+    # The session data would typically be cleaned up by the database teardown.
+
+@pytest.mark.asyncio
+async def test_chat_with_model_switch(http_client, session_id):
+    """Tests switching models within an existing session."""
+    print("\n--- Running test_chat_with_model_switch ---")
+    
+    # Send a message to the new session with a different model
+    payload_gemini = {"prompt": "What is the capital of France?", "model": "gemini"}
+    response_gemini = await http_client.post(f"/sessions/{session_id}/chat", json=payload_gemini)
+    assert response_gemini.status_code == 200
+    assert "Paris" in response_gemini.json()["answer"]
+    assert response_gemini.json()["model_used"] == "gemini"
+    print("✅ Chat (Model Switch to Gemini) test passed.")
+
+    # Switch back to the original model
+    payload_deepseek = {"prompt": "What is the largest ocean?", "model": "deepseek"}
+    response_deepseek = await http_client.post(f"/sessions/{session_id}/chat", json=payload_deepseek)
+    assert response_deepseek.status_code == 200
+    assert "Pacific Ocean" in response_deepseek.json()["answer"]
+    assert response_deepseek.json()["model_used"] == "deepseek"
+    print("✅ Chat (Model Switch back to DeepSeek) test passed.")
+
+@pytest.mark.asyncio
+async def test_chat_with_document_retrieval(http_client):
+    """
+    Tests injecting a document and using it for retrieval-augmented generation.
+    This test creates its own session and document for isolation.
+    """
+    print("\n--- Running test_chat_with_document_retrieval ---")
+    
+    # Create a new session for this RAG test
+    session_response = await http_client.post("/sessions", json={"user_id": "rag_tester", "model": "deepseek"})
+    assert session_response.status_code == 200
+    rag_session_id = session_response.json()["id"]
+    
+    # Add a new document with specific content for retrieval
+    doc_data = {"title": RAG_DOC_TITLE, "text": RAG_DOC_TEXT}
+    add_doc_response = await http_client.post("/documents", json=doc_data)
+    assert add_doc_response.status_code == 200
+    try:
+        message = add_doc_response.json().get("message", "")
+        rag_document_id = int(message.split(" with ID ")[-1])
+        print(f"Document for RAG created with ID: {rag_document_id}")
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    
+    try:
+        chat_payload = {
+            "prompt": RAG_PROMPT,
+            "document_id": rag_document_id,
+            "model": "deepseek",
+            "load_faiss_retriever": True
+        }
+        chat_response = await http_client.post(f"/sessions/{rag_session_id}/chat", json=chat_payload)
+        
+        assert chat_response.status_code == 200
+        chat_data = chat_response.json()
+        assert "Jane Doe" in chat_data["answer"]
+        assert "Nexus" in chat_data["answer"]
+        print("✅ Chat with document retrieval test passed.")
+    finally:
+        # Clean up the document after the test
+        delete_response = await http_client.delete(f"/documents/{rag_document_id}")
+        assert delete_response.status_code == 200
+        print(f"Document {rag_document_id} deleted successfully.")
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file
diff --git a/ai-hub/app/core/services/tts.py b/ai-hub/app/core/services/tts.py
new file mode 100644
index 0000000..c196795
--- /dev/null
+++ b/ai-hub/app/core/services/tts.py
@@ -0,0 +1,25 @@
+from typing import AsyncGenerator
+from app.core.providers.base import TTSProvider
+
+class TTSService:
+    """
+    Service class for generating speech from text using a TTS provider.
+    """
+    def __init__(self, tts_provider: TTSProvider):
+        """
+        Initializes the TTSService with a concrete TTS provider.
+        """
+        self.tts_provider = tts_provider
+
+    async def create_speech_stream(self, text: str) -> AsyncGenerator[bytes, None]:
+        """
+        Generates a stream of audio bytes from the given text using the configured
+        TTS provider.
+
+        Args:
+            text: The text to be converted to speech.
+
+        Returns:
+            An async generator that yields chunks of audio bytes.
+        """
+        return self.tts_provider.generate_speech(text)
\ No newline at end of file
diff --git a/ai-hub/integration_tests/conftest.py b/ai-hub/integration_tests/conftest.py
new file mode 100644
index 0000000..5d51010
--- /dev/null
+++ b/ai-hub/integration_tests/conftest.py
@@ -0,0 +1,57 @@
+import httpx
+import pytest_asyncio
+
+BASE_URL = "http://127.0.0.1:8000"
+
+@pytest_asyncio.fixture(scope="session")
+def base_url():
+    """Fixture to provide the base URL for the tests."""
+    return BASE_URL
+
+@pytest_asyncio.fixture(scope="function") # <-- Change scope to "function"
+async def http_client():
+    """
+    Fixture to provide an async HTTP client for all tests in the session.
+    A new client is created and closed properly using a try/finally block
+    to prevent "Event loop is closed" errors.
+    """
+    client = httpx.AsyncClient(base_url=BASE_URL, timeout=60.0)
+    try:
+        yield client
+    finally:
+        await client.aclose()
+
+@pytest_asyncio.fixture(scope="function")
+async def session_id(http_client):
+    """
+    Creates a new session before a test and cleans it up after.
+    Returns the session ID.
+    """
+    payload = {"user_id": "integration_tester", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    yield session_id
+    # No explicit session deletion is needed for this example,
+    # as sessions are typically managed by a database lifecycle.
+
+@pytest_asyncio.fixture(scope="function")
+async def document_id(http_client):
+    """
+    Creates a new document before a test and ensures it's deleted afterward.
+    Returns the document ID.
+    """
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    response = await http_client.post("/documents", json=doc_data)
+    assert response.status_code == 200
+    try:
+        message = response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+        
+    yield document_id
+    
+    # Teardown: Delete the document after the test
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_documents.py b/ai-hub/integration_tests/test_documents.py
new file mode 100644
index 0000000..c6d4362
--- /dev/null
+++ b/ai-hub/integration_tests/test_documents.py
@@ -0,0 +1,33 @@
+import pytest
+
+@pytest.mark.asyncio
+async def test_document_lifecycle(http_client):
+    """
+    Tests the full lifecycle of a document: add, list, and delete.
+    This is run as a single, sequential test for a clean state.
+    """
+    print("\n--- Running test_document_lifecycle ---")
+    
+    # 1. Add a new document
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    add_response = await http_client.post("/documents", json=doc_data)
+    assert add_response.status_code == 200
+    try:
+        message = add_response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    print(f"✅ Document for lifecycle test created with ID: {document_id}")
+    
+    # 2. List all documents and check if the new document is present
+    list_response = await http_client.get("/documents")
+    assert list_response.status_code == 200
+    ids_in_response = {doc["id"] for doc in list_response.json()["documents"]}
+    assert document_id in ids_in_response
+    print("✅ Document list test passed.")
+    
+    # 3. Delete the document
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
+    assert delete_response.json()["document_id"] == document_id
+    print("✅ Document delete test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_misc.py b/ai-hub/integration_tests/test_misc.py
new file mode 100644
index 0000000..3ff7925
--- /dev/null
+++ b/ai-hub/integration_tests/test_misc.py
@@ -0,0 +1,36 @@
+import pytest
+import httpx
+
+@pytest.mark.asyncio
+async def test_root_endpoint(http_client):
+    """
+    Tests if the root endpoint is alive and returns the correct status message.
+    """
+    print("\n--- Running test_root_endpoint ---")
+    response = await http_client.get("/")
+    assert response.status_code == 200
+    assert response.json() == {"status": "AI Model Hub is running!"}
+    print("✅ Root endpoint test passed.")
+
+# @pytest.mark.asyncio
+# async def test_create_speech_stream(http_client):
+#     """
+#     Tests the /speech endpoint for a successful audio stream response.
+#     """
+#     print("\n--- Running test_create_speech_stream ---")
+#     url = "/speech"
+#     payload = {"text": "Hello, world!"}
+
+#     # The `stream=True` parameter tells httpx to not read the entire response body
+#     # at once. We'll handle it manually to check for content.
+#     async with http_client.stream("POST", url, json=payload) as response:
+#         assert response.status_code == 200, f"Speech stream request failed. Response: {response.text}"
+#         assert response.headers.get("content-type") == "audio/wav"
+
+#         # Check that the response body is not empty by iterating over chunks.
+#         content_length = 0
+#         async for chunk in response.aiter_bytes():
+#             content_length += len(chunk)
+            
+#         assert content_length > 0
+#     print("✅ TTS stream test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_sessions.py b/ai-hub/integration_tests/test_sessions.py
new file mode 100644
index 0000000..44d63d7
--- /dev/null
+++ b/ai-hub/integration_tests/test_sessions.py
@@ -0,0 +1,107 @@
+import pytest
+
+# Test prompts and data
+CONTEXT_PROMPT = "Who is the CEO of Microsoft?"
+FOLLOW_UP_PROMPT = "When was he born?"
+RAG_DOC_TITLE = "Fictional Company History"
+RAG_DOC_TEXT = "The company AlphaCorp was founded in 2021 by Jane Doe. Their primary product is a smart home device called 'Nexus'."
+RAG_PROMPT = "Who founded AlphaCorp and what is their main product?"
+
+@pytest.mark.asyncio
+async def test_chat_in_session_lifecycle(http_client):
+    """
+    Tests a full session lifecycle from creation to conversational memory.
+    This test is a single, sequential unit.
+    """
+    print("\n--- Running test_chat_in_session_lifecycle ---")
+    
+    # 1. Create a new session
+    payload = {"user_id": "integration_tester_lifecycle", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    print(f"✅ Session created successfully with ID: {session_id}")
+    
+    # 2. First chat turn to establish context
+    chat_payload_1 = {"prompt": CONTEXT_PROMPT}
+    response_1 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_1)
+    assert response_1.status_code == 200
+    assert "Satya Nadella" in response_1.json()["answer"]
+    assert response_1.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 1 (context) test passed.")
+    
+    # 3. Second chat turn (follow-up) to test conversational memory
+    chat_payload_2 = {"prompt": FOLLOW_UP_PROMPT}
+    response_2 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_2)
+    assert response_2.status_code == 200
+    assert "1967" in response_2.json()["answer"]
+    assert response_2.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 2 (follow-up) test passed.")
+    
+    # 4. Cleanup (optional, but good practice if not using a test database that resets)
+    # The session data would typically be cleaned up by the database teardown.
+
+@pytest.mark.asyncio
+async def test_chat_with_model_switch(http_client, session_id):
+    """Tests switching models within an existing session."""
+    print("\n--- Running test_chat_with_model_switch ---")
+    
+    # Send a message to the new session with a different model
+    payload_gemini = {"prompt": "What is the capital of France?", "model": "gemini"}
+    response_gemini = await http_client.post(f"/sessions/{session_id}/chat", json=payload_gemini)
+    assert response_gemini.status_code == 200
+    assert "Paris" in response_gemini.json()["answer"]
+    assert response_gemini.json()["model_used"] == "gemini"
+    print("✅ Chat (Model Switch to Gemini) test passed.")
+
+    # Switch back to the original model
+    payload_deepseek = {"prompt": "What is the largest ocean?", "model": "deepseek"}
+    response_deepseek = await http_client.post(f"/sessions/{session_id}/chat", json=payload_deepseek)
+    assert response_deepseek.status_code == 200
+    assert "Pacific Ocean" in response_deepseek.json()["answer"]
+    assert response_deepseek.json()["model_used"] == "deepseek"
+    print("✅ Chat (Model Switch back to DeepSeek) test passed.")
+
+@pytest.mark.asyncio
+async def test_chat_with_document_retrieval(http_client):
+    """
+    Tests injecting a document and using it for retrieval-augmented generation.
+    This test creates its own session and document for isolation.
+    """
+    print("\n--- Running test_chat_with_document_retrieval ---")
+    
+    # Create a new session for this RAG test
+    session_response = await http_client.post("/sessions", json={"user_id": "rag_tester", "model": "deepseek"})
+    assert session_response.status_code == 200
+    rag_session_id = session_response.json()["id"]
+    
+    # Add a new document with specific content for retrieval
+    doc_data = {"title": RAG_DOC_TITLE, "text": RAG_DOC_TEXT}
+    add_doc_response = await http_client.post("/documents", json=doc_data)
+    assert add_doc_response.status_code == 200
+    try:
+        message = add_doc_response.json().get("message", "")
+        rag_document_id = int(message.split(" with ID ")[-1])
+        print(f"Document for RAG created with ID: {rag_document_id}")
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    
+    try:
+        chat_payload = {
+            "prompt": RAG_PROMPT,
+            "document_id": rag_document_id,
+            "model": "deepseek",
+            "load_faiss_retriever": True
+        }
+        chat_response = await http_client.post(f"/sessions/{rag_session_id}/chat", json=chat_payload)
+        
+        assert chat_response.status_code == 200
+        chat_data = chat_response.json()
+        assert "Jane Doe" in chat_data["answer"]
+        assert "Nexus" in chat_data["answer"]
+        print("✅ Chat with document retrieval test passed.")
+    finally:
+        # Clean up the document after the test
+        delete_response = await http_client.delete(f"/documents/{rag_document_id}")
+        assert delete_response.status_code == 200
+        print(f"Document {rag_document_id} deleted successfully.")
\ No newline at end of file
diff --git a/ai-hub/run_integration_tests.sh b/ai-hub/run_integration_tests.sh
index 65062f6..ef45dc0 100644
--- a/ai-hub/run_integration_tests.sh
+++ b/ai-hub/run_integration_tests.sh
@@ -4,15 +4,45 @@
 # It starts the FastAPI server, runs the specified tests, and then shuts down the server.
 
 # --- Configuration ---
-# Set the default path for tests to run. This will be used if no argument is provided.
-DEFAULT_TEST_PATH="integration_tests/"
-# You can override the default with a command-line argument, e.g., './run_integration_tests.sh tests/test_app.py'
-TEST_PATH=${1:-$DEFAULT_TEST_PATH}
+# You can define aliases for your test file paths here.
+TEST_SUITES=(
+  "All tests"
+  "integration_tests/test_sessions.py"
+  "integration_tests/test_documents.py"
+  "integration_tests/test_misc.py"
+)
+TEST_PATHS=(
+  "integration_tests/"
+  "integration_tests/test_sessions.py"
+  "integration_tests/test_documents.py"
+  "integration_tests/test_misc.py"
+)
 
 export DB_MODE=sqlite
 export LOCAL_DB_PATH="data/integration_test_ai_hub.db"
 export FAISS_INDEX_PATH="data/integration_test_faiss_index.bin"
 
+# --- User Interaction ---
+echo "--- AI Hub Test Runner ---"
+echo "Select a test suite to run:"
+
+# Display the menu options
+for i in "${!TEST_SUITES[@]}"; do
+  echo "$((i+1)). ${TEST_SUITES[$i]}"
+done
+
+# Prompt for user input
+read -p "Enter the number of your choice (1-${#TEST_SUITES[@]}): " choice
+
+# Validate input and set the TEST_PATH
+if [[ "$choice" =~ ^[0-9]+$ ]] && [ "$choice" -ge 1 ] && [ "$choice" -le "${#TEST_SUITES[@]}" ]; then
+  TEST_PATH=${TEST_PATHS[$((choice-1))]}
+  echo "You have selected: ${TEST_SUITES[$((choice-1))]}"
+else
+  echo "Invalid choice. Running all tests by default."
+  TEST_PATH=${TEST_PATHS[0]}
+fi
+
 # --- Pre-test Cleanup ---
 # Check for and remove old test files to ensure a clean test environment.
 echo "--- Checking for and removing old test files ---"
@@ -62,4 +92,4 @@
 
 # The 'trap' will automatically call the cleanup function now.
 # Exit with the same code as the test script (0 for success, non-zero for failure).
-exit $TEST_EXIT_CODE
+exit $TEST_EXIT_CODE
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file
diff --git a/ai-hub/app/core/services/tts.py b/ai-hub/app/core/services/tts.py
new file mode 100644
index 0000000..c196795
--- /dev/null
+++ b/ai-hub/app/core/services/tts.py
@@ -0,0 +1,25 @@
+from typing import AsyncGenerator
+from app.core.providers.base import TTSProvider
+
+class TTSService:
+    """
+    Service class for generating speech from text using a TTS provider.
+    """
+    def __init__(self, tts_provider: TTSProvider):
+        """
+        Initializes the TTSService with a concrete TTS provider.
+        """
+        self.tts_provider = tts_provider
+
+    async def create_speech_stream(self, text: str) -> AsyncGenerator[bytes, None]:
+        """
+        Generates a stream of audio bytes from the given text using the configured
+        TTS provider.
+
+        Args:
+            text: The text to be converted to speech.
+
+        Returns:
+            An async generator that yields chunks of audio bytes.
+        """
+        return self.tts_provider.generate_speech(text)
\ No newline at end of file
diff --git a/ai-hub/integration_tests/conftest.py b/ai-hub/integration_tests/conftest.py
new file mode 100644
index 0000000..5d51010
--- /dev/null
+++ b/ai-hub/integration_tests/conftest.py
@@ -0,0 +1,57 @@
+import httpx
+import pytest_asyncio
+
+BASE_URL = "http://127.0.0.1:8000"
+
+@pytest_asyncio.fixture(scope="session")
+def base_url():
+    """Fixture to provide the base URL for the tests."""
+    return BASE_URL
+
+@pytest_asyncio.fixture(scope="function") # <-- Change scope to "function"
+async def http_client():
+    """
+    Fixture to provide an async HTTP client for all tests in the session.
+    A new client is created and closed properly using a try/finally block
+    to prevent "Event loop is closed" errors.
+    """
+    client = httpx.AsyncClient(base_url=BASE_URL, timeout=60.0)
+    try:
+        yield client
+    finally:
+        await client.aclose()
+
+@pytest_asyncio.fixture(scope="function")
+async def session_id(http_client):
+    """
+    Creates a new session before a test and cleans it up after.
+    Returns the session ID.
+    """
+    payload = {"user_id": "integration_tester", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    yield session_id
+    # No explicit session deletion is needed for this example,
+    # as sessions are typically managed by a database lifecycle.
+
+@pytest_asyncio.fixture(scope="function")
+async def document_id(http_client):
+    """
+    Creates a new document before a test and ensures it's deleted afterward.
+    Returns the document ID.
+    """
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    response = await http_client.post("/documents", json=doc_data)
+    assert response.status_code == 200
+    try:
+        message = response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+        
+    yield document_id
+    
+    # Teardown: Delete the document after the test
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_documents.py b/ai-hub/integration_tests/test_documents.py
new file mode 100644
index 0000000..c6d4362
--- /dev/null
+++ b/ai-hub/integration_tests/test_documents.py
@@ -0,0 +1,33 @@
+import pytest
+
+@pytest.mark.asyncio
+async def test_document_lifecycle(http_client):
+    """
+    Tests the full lifecycle of a document: add, list, and delete.
+    This is run as a single, sequential test for a clean state.
+    """
+    print("\n--- Running test_document_lifecycle ---")
+    
+    # 1. Add a new document
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    add_response = await http_client.post("/documents", json=doc_data)
+    assert add_response.status_code == 200
+    try:
+        message = add_response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    print(f"✅ Document for lifecycle test created with ID: {document_id}")
+    
+    # 2. List all documents and check if the new document is present
+    list_response = await http_client.get("/documents")
+    assert list_response.status_code == 200
+    ids_in_response = {doc["id"] for doc in list_response.json()["documents"]}
+    assert document_id in ids_in_response
+    print("✅ Document list test passed.")
+    
+    # 3. Delete the document
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
+    assert delete_response.json()["document_id"] == document_id
+    print("✅ Document delete test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_misc.py b/ai-hub/integration_tests/test_misc.py
new file mode 100644
index 0000000..3ff7925
--- /dev/null
+++ b/ai-hub/integration_tests/test_misc.py
@@ -0,0 +1,36 @@
+import pytest
+import httpx
+
+@pytest.mark.asyncio
+async def test_root_endpoint(http_client):
+    """
+    Tests if the root endpoint is alive and returns the correct status message.
+    """
+    print("\n--- Running test_root_endpoint ---")
+    response = await http_client.get("/")
+    assert response.status_code == 200
+    assert response.json() == {"status": "AI Model Hub is running!"}
+    print("✅ Root endpoint test passed.")
+
+# @pytest.mark.asyncio
+# async def test_create_speech_stream(http_client):
+#     """
+#     Tests the /speech endpoint for a successful audio stream response.
+#     """
+#     print("\n--- Running test_create_speech_stream ---")
+#     url = "/speech"
+#     payload = {"text": "Hello, world!"}
+
+#     # The `stream=True` parameter tells httpx to not read the entire response body
+#     # at once. We'll handle it manually to check for content.
+#     async with http_client.stream("POST", url, json=payload) as response:
+#         assert response.status_code == 200, f"Speech stream request failed. Response: {response.text}"
+#         assert response.headers.get("content-type") == "audio/wav"
+
+#         # Check that the response body is not empty by iterating over chunks.
+#         content_length = 0
+#         async for chunk in response.aiter_bytes():
+#             content_length += len(chunk)
+            
+#         assert content_length > 0
+#     print("✅ TTS stream test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_sessions.py b/ai-hub/integration_tests/test_sessions.py
new file mode 100644
index 0000000..44d63d7
--- /dev/null
+++ b/ai-hub/integration_tests/test_sessions.py
@@ -0,0 +1,107 @@
+import pytest
+
+# Test prompts and data
+CONTEXT_PROMPT = "Who is the CEO of Microsoft?"
+FOLLOW_UP_PROMPT = "When was he born?"
+RAG_DOC_TITLE = "Fictional Company History"
+RAG_DOC_TEXT = "The company AlphaCorp was founded in 2021 by Jane Doe. Their primary product is a smart home device called 'Nexus'."
+RAG_PROMPT = "Who founded AlphaCorp and what is their main product?"
+
+@pytest.mark.asyncio
+async def test_chat_in_session_lifecycle(http_client):
+    """
+    Tests a full session lifecycle from creation to conversational memory.
+    This test is a single, sequential unit.
+    """
+    print("\n--- Running test_chat_in_session_lifecycle ---")
+    
+    # 1. Create a new session
+    payload = {"user_id": "integration_tester_lifecycle", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    print(f"✅ Session created successfully with ID: {session_id}")
+    
+    # 2. First chat turn to establish context
+    chat_payload_1 = {"prompt": CONTEXT_PROMPT}
+    response_1 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_1)
+    assert response_1.status_code == 200
+    assert "Satya Nadella" in response_1.json()["answer"]
+    assert response_1.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 1 (context) test passed.")
+    
+    # 3. Second chat turn (follow-up) to test conversational memory
+    chat_payload_2 = {"prompt": FOLLOW_UP_PROMPT}
+    response_2 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_2)
+    assert response_2.status_code == 200
+    assert "1967" in response_2.json()["answer"]
+    assert response_2.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 2 (follow-up) test passed.")
+    
+    # 4. Cleanup (optional, but good practice if not using a test database that resets)
+    # The session data would typically be cleaned up by the database teardown.
+
+@pytest.mark.asyncio
+async def test_chat_with_model_switch(http_client, session_id):
+    """Tests switching models within an existing session."""
+    print("\n--- Running test_chat_with_model_switch ---")
+    
+    # Send a message to the new session with a different model
+    payload_gemini = {"prompt": "What is the capital of France?", "model": "gemini"}
+    response_gemini = await http_client.post(f"/sessions/{session_id}/chat", json=payload_gemini)
+    assert response_gemini.status_code == 200
+    assert "Paris" in response_gemini.json()["answer"]
+    assert response_gemini.json()["model_used"] == "gemini"
+    print("✅ Chat (Model Switch to Gemini) test passed.")
+
+    # Switch back to the original model
+    payload_deepseek = {"prompt": "What is the largest ocean?", "model": "deepseek"}
+    response_deepseek = await http_client.post(f"/sessions/{session_id}/chat", json=payload_deepseek)
+    assert response_deepseek.status_code == 200
+    assert "Pacific Ocean" in response_deepseek.json()["answer"]
+    assert response_deepseek.json()["model_used"] == "deepseek"
+    print("✅ Chat (Model Switch back to DeepSeek) test passed.")
+
+@pytest.mark.asyncio
+async def test_chat_with_document_retrieval(http_client):
+    """
+    Tests injecting a document and using it for retrieval-augmented generation.
+    This test creates its own session and document for isolation.
+    """
+    print("\n--- Running test_chat_with_document_retrieval ---")
+    
+    # Create a new session for this RAG test
+    session_response = await http_client.post("/sessions", json={"user_id": "rag_tester", "model": "deepseek"})
+    assert session_response.status_code == 200
+    rag_session_id = session_response.json()["id"]
+    
+    # Add a new document with specific content for retrieval
+    doc_data = {"title": RAG_DOC_TITLE, "text": RAG_DOC_TEXT}
+    add_doc_response = await http_client.post("/documents", json=doc_data)
+    assert add_doc_response.status_code == 200
+    try:
+        message = add_doc_response.json().get("message", "")
+        rag_document_id = int(message.split(" with ID ")[-1])
+        print(f"Document for RAG created with ID: {rag_document_id}")
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    
+    try:
+        chat_payload = {
+            "prompt": RAG_PROMPT,
+            "document_id": rag_document_id,
+            "model": "deepseek",
+            "load_faiss_retriever": True
+        }
+        chat_response = await http_client.post(f"/sessions/{rag_session_id}/chat", json=chat_payload)
+        
+        assert chat_response.status_code == 200
+        chat_data = chat_response.json()
+        assert "Jane Doe" in chat_data["answer"]
+        assert "Nexus" in chat_data["answer"]
+        print("✅ Chat with document retrieval test passed.")
+    finally:
+        # Clean up the document after the test
+        delete_response = await http_client.delete(f"/documents/{rag_document_id}")
+        assert delete_response.status_code == 200
+        print(f"Document {rag_document_id} deleted successfully.")
\ No newline at end of file
diff --git a/ai-hub/run_integration_tests.sh b/ai-hub/run_integration_tests.sh
index 65062f6..ef45dc0 100644
--- a/ai-hub/run_integration_tests.sh
+++ b/ai-hub/run_integration_tests.sh
@@ -4,15 +4,45 @@
 # It starts the FastAPI server, runs the specified tests, and then shuts down the server.
 
 # --- Configuration ---
-# Set the default path for tests to run. This will be used if no argument is provided.
-DEFAULT_TEST_PATH="integration_tests/"
-# You can override the default with a command-line argument, e.g., './run_integration_tests.sh tests/test_app.py'
-TEST_PATH=${1:-$DEFAULT_TEST_PATH}
+# You can define aliases for your test file paths here.
+TEST_SUITES=(
+  "All tests"
+  "integration_tests/test_sessions.py"
+  "integration_tests/test_documents.py"
+  "integration_tests/test_misc.py"
+)
+TEST_PATHS=(
+  "integration_tests/"
+  "integration_tests/test_sessions.py"
+  "integration_tests/test_documents.py"
+  "integration_tests/test_misc.py"
+)
 
 export DB_MODE=sqlite
 export LOCAL_DB_PATH="data/integration_test_ai_hub.db"
 export FAISS_INDEX_PATH="data/integration_test_faiss_index.bin"
 
+# --- User Interaction ---
+echo "--- AI Hub Test Runner ---"
+echo "Select a test suite to run:"
+
+# Display the menu options
+for i in "${!TEST_SUITES[@]}"; do
+  echo "$((i+1)). ${TEST_SUITES[$i]}"
+done
+
+# Prompt for user input
+read -p "Enter the number of your choice (1-${#TEST_SUITES[@]}): " choice
+
+# Validate input and set the TEST_PATH
+if [[ "$choice" =~ ^[0-9]+$ ]] && [ "$choice" -ge 1 ] && [ "$choice" -le "${#TEST_SUITES[@]}" ]; then
+  TEST_PATH=${TEST_PATHS[$((choice-1))]}
+  echo "You have selected: ${TEST_SUITES[$((choice-1))]}"
+else
+  echo "Invalid choice. Running all tests by default."
+  TEST_PATH=${TEST_PATHS[0]}
+fi
+
 # --- Pre-test Cleanup ---
 # Check for and remove old test files to ensure a clean test environment.
 echo "--- Checking for and removing old test files ---"
@@ -62,4 +92,4 @@
 
 # The 'trap' will automatically call the cleanup function now.
 # Exit with the same code as the test script (0 for success, non-zero for failure).
-exit $TEST_EXIT_CODE
+exit $TEST_EXIT_CODE
\ No newline at end of file
diff --git a/ai-hub/tests/api/test_dependencies.py b/ai-hub/tests/api/test_dependencies.py
index 65490e8..d2e0472 100644
--- a/ai-hub/tests/api/test_dependencies.py
+++ b/ai-hub/tests/api/test_dependencies.py
@@ -9,6 +9,7 @@
 from app.api.dependencies import get_db, get_current_user, ServiceContainer
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService # Added this import
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.retrievers.base_retriever import Retriever
 
@@ -97,11 +98,13 @@
     # The DocumentService constructor needs a .embedder attribute on the vector_store
     mock_vector_store.embedder = MagicMock()
     mock_retrievers = [MagicMock(spec=Retriever), MagicMock(spec=Retriever)]
+    mock_tts_service = MagicMock(spec=TTSService)
 
     # Act: Instantiate the ServiceContainer
     container = ServiceContainer(
         vector_store=mock_vector_store,
-        retrievers=mock_retrievers
+        retrievers=mock_retrievers,
+        tts_service=mock_tts_service # Passing the mock TTS service
     )
 
     # Assert: Check if the services were created and configured correctly
@@ -110,3 +113,7 @@
     
     assert isinstance(container.rag_service, RAGService)
     assert container.rag_service.retrievers == mock_retrievers
+    
+    # Assert for the tts_service as well
+    assert isinstance(container.tts_service, TTSService)
+    assert container.tts_service == mock_tts_service

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file
diff --git a/ai-hub/app/core/services/tts.py b/ai-hub/app/core/services/tts.py
new file mode 100644
index 0000000..c196795
--- /dev/null
+++ b/ai-hub/app/core/services/tts.py
@@ -0,0 +1,25 @@
+from typing import AsyncGenerator
+from app.core.providers.base import TTSProvider
+
+class TTSService:
+    """
+    Service class for generating speech from text using a TTS provider.
+    """
+    def __init__(self, tts_provider: TTSProvider):
+        """
+        Initializes the TTSService with a concrete TTS provider.
+        """
+        self.tts_provider = tts_provider
+
+    async def create_speech_stream(self, text: str) -> AsyncGenerator[bytes, None]:
+        """
+        Generates a stream of audio bytes from the given text using the configured
+        TTS provider.
+
+        Args:
+            text: The text to be converted to speech.
+
+        Returns:
+            An async generator that yields chunks of audio bytes.
+        """
+        return self.tts_provider.generate_speech(text)
\ No newline at end of file
diff --git a/ai-hub/integration_tests/conftest.py b/ai-hub/integration_tests/conftest.py
new file mode 100644
index 0000000..5d51010
--- /dev/null
+++ b/ai-hub/integration_tests/conftest.py
@@ -0,0 +1,57 @@
+import httpx
+import pytest_asyncio
+
+BASE_URL = "http://127.0.0.1:8000"
+
+@pytest_asyncio.fixture(scope="session")
+def base_url():
+    """Fixture to provide the base URL for the tests."""
+    return BASE_URL
+
+@pytest_asyncio.fixture(scope="function") # <-- Change scope to "function"
+async def http_client():
+    """
+    Fixture to provide an async HTTP client for all tests in the session.
+    A new client is created and closed properly using a try/finally block
+    to prevent "Event loop is closed" errors.
+    """
+    client = httpx.AsyncClient(base_url=BASE_URL, timeout=60.0)
+    try:
+        yield client
+    finally:
+        await client.aclose()
+
+@pytest_asyncio.fixture(scope="function")
+async def session_id(http_client):
+    """
+    Creates a new session before a test and cleans it up after.
+    Returns the session ID.
+    """
+    payload = {"user_id": "integration_tester", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    yield session_id
+    # No explicit session deletion is needed for this example,
+    # as sessions are typically managed by a database lifecycle.
+
+@pytest_asyncio.fixture(scope="function")
+async def document_id(http_client):
+    """
+    Creates a new document before a test and ensures it's deleted afterward.
+    Returns the document ID.
+    """
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    response = await http_client.post("/documents", json=doc_data)
+    assert response.status_code == 200
+    try:
+        message = response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+        
+    yield document_id
+    
+    # Teardown: Delete the document after the test
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_documents.py b/ai-hub/integration_tests/test_documents.py
new file mode 100644
index 0000000..c6d4362
--- /dev/null
+++ b/ai-hub/integration_tests/test_documents.py
@@ -0,0 +1,33 @@
+import pytest
+
+@pytest.mark.asyncio
+async def test_document_lifecycle(http_client):
+    """
+    Tests the full lifecycle of a document: add, list, and delete.
+    This is run as a single, sequential test for a clean state.
+    """
+    print("\n--- Running test_document_lifecycle ---")
+    
+    # 1. Add a new document
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    add_response = await http_client.post("/documents", json=doc_data)
+    assert add_response.status_code == 200
+    try:
+        message = add_response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    print(f"✅ Document for lifecycle test created with ID: {document_id}")
+    
+    # 2. List all documents and check if the new document is present
+    list_response = await http_client.get("/documents")
+    assert list_response.status_code == 200
+    ids_in_response = {doc["id"] for doc in list_response.json()["documents"]}
+    assert document_id in ids_in_response
+    print("✅ Document list test passed.")
+    
+    # 3. Delete the document
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
+    assert delete_response.json()["document_id"] == document_id
+    print("✅ Document delete test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_misc.py b/ai-hub/integration_tests/test_misc.py
new file mode 100644
index 0000000..3ff7925
--- /dev/null
+++ b/ai-hub/integration_tests/test_misc.py
@@ -0,0 +1,36 @@
+import pytest
+import httpx
+
+@pytest.mark.asyncio
+async def test_root_endpoint(http_client):
+    """
+    Tests if the root endpoint is alive and returns the correct status message.
+    """
+    print("\n--- Running test_root_endpoint ---")
+    response = await http_client.get("/")
+    assert response.status_code == 200
+    assert response.json() == {"status": "AI Model Hub is running!"}
+    print("✅ Root endpoint test passed.")
+
+# @pytest.mark.asyncio
+# async def test_create_speech_stream(http_client):
+#     """
+#     Tests the /speech endpoint for a successful audio stream response.
+#     """
+#     print("\n--- Running test_create_speech_stream ---")
+#     url = "/speech"
+#     payload = {"text": "Hello, world!"}
+
+#     # The `stream=True` parameter tells httpx to not read the entire response body
+#     # at once. We'll handle it manually to check for content.
+#     async with http_client.stream("POST", url, json=payload) as response:
+#         assert response.status_code == 200, f"Speech stream request failed. Response: {response.text}"
+#         assert response.headers.get("content-type") == "audio/wav"
+
+#         # Check that the response body is not empty by iterating over chunks.
+#         content_length = 0
+#         async for chunk in response.aiter_bytes():
+#             content_length += len(chunk)
+            
+#         assert content_length > 0
+#     print("✅ TTS stream test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_sessions.py b/ai-hub/integration_tests/test_sessions.py
new file mode 100644
index 0000000..44d63d7
--- /dev/null
+++ b/ai-hub/integration_tests/test_sessions.py
@@ -0,0 +1,107 @@
+import pytest
+
+# Test prompts and data
+CONTEXT_PROMPT = "Who is the CEO of Microsoft?"
+FOLLOW_UP_PROMPT = "When was he born?"
+RAG_DOC_TITLE = "Fictional Company History"
+RAG_DOC_TEXT = "The company AlphaCorp was founded in 2021 by Jane Doe. Their primary product is a smart home device called 'Nexus'."
+RAG_PROMPT = "Who founded AlphaCorp and what is their main product?"
+
+@pytest.mark.asyncio
+async def test_chat_in_session_lifecycle(http_client):
+    """
+    Tests a full session lifecycle from creation to conversational memory.
+    This test is a single, sequential unit.
+    """
+    print("\n--- Running test_chat_in_session_lifecycle ---")
+    
+    # 1. Create a new session
+    payload = {"user_id": "integration_tester_lifecycle", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    print(f"✅ Session created successfully with ID: {session_id}")
+    
+    # 2. First chat turn to establish context
+    chat_payload_1 = {"prompt": CONTEXT_PROMPT}
+    response_1 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_1)
+    assert response_1.status_code == 200
+    assert "Satya Nadella" in response_1.json()["answer"]
+    assert response_1.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 1 (context) test passed.")
+    
+    # 3. Second chat turn (follow-up) to test conversational memory
+    chat_payload_2 = {"prompt": FOLLOW_UP_PROMPT}
+    response_2 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_2)
+    assert response_2.status_code == 200
+    assert "1967" in response_2.json()["answer"]
+    assert response_2.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 2 (follow-up) test passed.")
+    
+    # 4. Cleanup (optional, but good practice if not using a test database that resets)
+    # The session data would typically be cleaned up by the database teardown.
+
+@pytest.mark.asyncio
+async def test_chat_with_model_switch(http_client, session_id):
+    """Tests switching models within an existing session."""
+    print("\n--- Running test_chat_with_model_switch ---")
+    
+    # Send a message to the new session with a different model
+    payload_gemini = {"prompt": "What is the capital of France?", "model": "gemini"}
+    response_gemini = await http_client.post(f"/sessions/{session_id}/chat", json=payload_gemini)
+    assert response_gemini.status_code == 200
+    assert "Paris" in response_gemini.json()["answer"]
+    assert response_gemini.json()["model_used"] == "gemini"
+    print("✅ Chat (Model Switch to Gemini) test passed.")
+
+    # Switch back to the original model
+    payload_deepseek = {"prompt": "What is the largest ocean?", "model": "deepseek"}
+    response_deepseek = await http_client.post(f"/sessions/{session_id}/chat", json=payload_deepseek)
+    assert response_deepseek.status_code == 200
+    assert "Pacific Ocean" in response_deepseek.json()["answer"]
+    assert response_deepseek.json()["model_used"] == "deepseek"
+    print("✅ Chat (Model Switch back to DeepSeek) test passed.")
+
+@pytest.mark.asyncio
+async def test_chat_with_document_retrieval(http_client):
+    """
+    Tests injecting a document and using it for retrieval-augmented generation.
+    This test creates its own session and document for isolation.
+    """
+    print("\n--- Running test_chat_with_document_retrieval ---")
+    
+    # Create a new session for this RAG test
+    session_response = await http_client.post("/sessions", json={"user_id": "rag_tester", "model": "deepseek"})
+    assert session_response.status_code == 200
+    rag_session_id = session_response.json()["id"]
+    
+    # Add a new document with specific content for retrieval
+    doc_data = {"title": RAG_DOC_TITLE, "text": RAG_DOC_TEXT}
+    add_doc_response = await http_client.post("/documents", json=doc_data)
+    assert add_doc_response.status_code == 200
+    try:
+        message = add_doc_response.json().get("message", "")
+        rag_document_id = int(message.split(" with ID ")[-1])
+        print(f"Document for RAG created with ID: {rag_document_id}")
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    
+    try:
+        chat_payload = {
+            "prompt": RAG_PROMPT,
+            "document_id": rag_document_id,
+            "model": "deepseek",
+            "load_faiss_retriever": True
+        }
+        chat_response = await http_client.post(f"/sessions/{rag_session_id}/chat", json=chat_payload)
+        
+        assert chat_response.status_code == 200
+        chat_data = chat_response.json()
+        assert "Jane Doe" in chat_data["answer"]
+        assert "Nexus" in chat_data["answer"]
+        print("✅ Chat with document retrieval test passed.")
+    finally:
+        # Clean up the document after the test
+        delete_response = await http_client.delete(f"/documents/{rag_document_id}")
+        assert delete_response.status_code == 200
+        print(f"Document {rag_document_id} deleted successfully.")
\ No newline at end of file
diff --git a/ai-hub/run_integration_tests.sh b/ai-hub/run_integration_tests.sh
index 65062f6..ef45dc0 100644
--- a/ai-hub/run_integration_tests.sh
+++ b/ai-hub/run_integration_tests.sh
@@ -4,15 +4,45 @@
 # It starts the FastAPI server, runs the specified tests, and then shuts down the server.
 
 # --- Configuration ---
-# Set the default path for tests to run. This will be used if no argument is provided.
-DEFAULT_TEST_PATH="integration_tests/"
-# You can override the default with a command-line argument, e.g., './run_integration_tests.sh tests/test_app.py'
-TEST_PATH=${1:-$DEFAULT_TEST_PATH}
+# You can define aliases for your test file paths here.
+TEST_SUITES=(
+  "All tests"
+  "integration_tests/test_sessions.py"
+  "integration_tests/test_documents.py"
+  "integration_tests/test_misc.py"
+)
+TEST_PATHS=(
+  "integration_tests/"
+  "integration_tests/test_sessions.py"
+  "integration_tests/test_documents.py"
+  "integration_tests/test_misc.py"
+)
 
 export DB_MODE=sqlite
 export LOCAL_DB_PATH="data/integration_test_ai_hub.db"
 export FAISS_INDEX_PATH="data/integration_test_faiss_index.bin"
 
+# --- User Interaction ---
+echo "--- AI Hub Test Runner ---"
+echo "Select a test suite to run:"
+
+# Display the menu options
+for i in "${!TEST_SUITES[@]}"; do
+  echo "$((i+1)). ${TEST_SUITES[$i]}"
+done
+
+# Prompt for user input
+read -p "Enter the number of your choice (1-${#TEST_SUITES[@]}): " choice
+
+# Validate input and set the TEST_PATH
+if [[ "$choice" =~ ^[0-9]+$ ]] && [ "$choice" -ge 1 ] && [ "$choice" -le "${#TEST_SUITES[@]}" ]; then
+  TEST_PATH=${TEST_PATHS[$((choice-1))]}
+  echo "You have selected: ${TEST_SUITES[$((choice-1))]}"
+else
+  echo "Invalid choice. Running all tests by default."
+  TEST_PATH=${TEST_PATHS[0]}
+fi
+
 # --- Pre-test Cleanup ---
 # Check for and remove old test files to ensure a clean test environment.
 echo "--- Checking for and removing old test files ---"
@@ -62,4 +92,4 @@
 
 # The 'trap' will automatically call the cleanup function now.
 # Exit with the same code as the test script (0 for success, non-zero for failure).
-exit $TEST_EXIT_CODE
+exit $TEST_EXIT_CODE
\ No newline at end of file
diff --git a/ai-hub/tests/api/test_dependencies.py b/ai-hub/tests/api/test_dependencies.py
index 65490e8..d2e0472 100644
--- a/ai-hub/tests/api/test_dependencies.py
+++ b/ai-hub/tests/api/test_dependencies.py
@@ -9,6 +9,7 @@
 from app.api.dependencies import get_db, get_current_user, ServiceContainer
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService # Added this import
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.retrievers.base_retriever import Retriever
 
@@ -97,11 +98,13 @@
     # The DocumentService constructor needs a .embedder attribute on the vector_store
     mock_vector_store.embedder = MagicMock()
     mock_retrievers = [MagicMock(spec=Retriever), MagicMock(spec=Retriever)]
+    mock_tts_service = MagicMock(spec=TTSService)
 
     # Act: Instantiate the ServiceContainer
     container = ServiceContainer(
         vector_store=mock_vector_store,
-        retrievers=mock_retrievers
+        retrievers=mock_retrievers,
+        tts_service=mock_tts_service # Passing the mock TTS service
     )
 
     # Assert: Check if the services were created and configured correctly
@@ -110,3 +113,7 @@
     
     assert isinstance(container.rag_service, RAGService)
     assert container.rag_service.retrievers == mock_retrievers
+    
+    # Assert for the tts_service as well
+    assert isinstance(container.tts_service, TTSService)
+    assert container.tts_service == mock_tts_service
diff --git a/ai-hub/tests/api/test_routes.py b/ai-hub/tests/api/test_routes.py
index 8d841a2..6d5cd3e 100644
--- a/ai-hub/tests/api/test_routes.py
+++ b/ai-hub/tests/api/test_routes.py
@@ -1,15 +1,18 @@
-# tests/app/api/test_routes.py
+# tests/api/test_routes.py
 import pytest
 from unittest.mock import MagicMock, AsyncMock
 from fastapi import FastAPI
 from fastapi.testclient import TestClient
 from sqlalchemy.orm import Session
 from datetime import datetime
+from httpx import AsyncClient, ASGITransport
+
 
 # Import the dependencies and router factory
 from app.api.dependencies import get_db, ServiceContainer
 from app.core.services.rag import RAGService
 from app.core.services.document import DocumentService
+from app.core.services.tts import TTSService 
 from app.api.routes import create_api_router
 from app.db import models # Import your SQLAlchemy models
 
@@ -24,11 +27,15 @@
     # Mock individual services
     mock_rag_service = MagicMock(spec=RAGService)
     mock_document_service = MagicMock(spec=DocumentService)
-
+    
+    # Use AsyncMock for the TTS service since its methods are async
+    mock_tts_service = MagicMock(spec=TTSService)
+    
     # Create a mock ServiceContainer that holds the mocked services
     mock_services = MagicMock(spec=ServiceContainer)
     mock_services.rag_service = mock_rag_service
     mock_services.document_service = mock_document_service
+    mock_services.tts_service = mock_tts_service
     
     # Mock the database session
     mock_db_session = MagicMock(spec=Session)
@@ -80,8 +87,6 @@
     assert response.status_code == 200
     assert response.json() == {"answer": "Mocked response", "model_used": "deepseek"}
     
-    # Verify that chat_with_rag was called with the default model 'deepseek'
-    # and the default load_faiss_retriever=False
     mock_services.rag_service.chat_with_rag.assert_called_once_with(
         db=mock_services.rag_service.chat_with_rag.call_args.kwargs['db'],
         session_id=42,
@@ -102,13 +107,12 @@
     assert response.status_code == 200
     assert response.json() == {"answer": "Mocked response from Gemini", "model_used": "gemini"}
     
-    # Verify that chat_with_rag was called with the specified model 'gemini'
     mock_services.rag_service.chat_with_rag.assert_called_once_with(
         db=mock_services.rag_service.chat_with_rag.call_args.kwargs['db'],
         session_id=42,
         prompt="Hello there, Gemini!",
         model="gemini",
-        load_faiss_retriever=False # It should still default to False
+        load_faiss_retriever=False
     )
 
 def test_chat_in_session_with_faiss_retriever(client):
@@ -126,29 +130,25 @@
     assert response.status_code == 200
     assert response.json() == {"answer": "Response with context", "model_used": "deepseek"}
     
-    # Verify that chat_with_rag was called with the correct parameters
     mock_services.rag_service.chat_with_rag.assert_called_once_with(
         db=mock_services.rag_service.chat_with_rag.call_args.kwargs['db'],
         session_id=42,
         prompt="What is RAG?",
-        model="deepseek", # The model still defaults to deepseek
-        load_faiss_retriever=True # Verify that the retriever was explicitly enabled
+        model="deepseek",
+        load_faiss_retriever=True
     )
 
 def test_get_session_messages_success(client):
     """Tests retrieving the message history for a session."""
     test_client, mock_services = client
-    # Arrange: Mock the service to return a list of message objects
     mock_history = [
         models.Message(sender="user", content="Hello", created_at=datetime.now()),
         models.Message(sender="assistant", content="Hi there!", created_at=datetime.now())
     ]
     mock_services.rag_service.get_message_history.return_value = mock_history
     
-    # Act
     response = test_client.get("/sessions/123/messages")
     
-    # Assert
     assert response.status_code == 200
     response_data = response.json()
     assert response_data["session_id"] == 123
@@ -163,13 +163,10 @@
 def test_get_session_messages_not_found(client):
     """Tests retrieving messages for a session that does not exist."""
     test_client, mock_services = client
-    # Arrange: Mock the service to return None, indicating the session wasn't found
     mock_services.rag_service.get_message_history.return_value = None
     
-    # Act
     response = test_client.get("/sessions/999/messages")
     
-    # Assert
     assert response.status_code == 404
     assert response.json()["detail"] == "Session with ID 999 not found."
 
@@ -210,3 +207,34 @@
     mock_services.document_service.delete_document.return_value = None
     response = test_client.delete("/documents/999")
     assert response.status_code == 404
+    
+# --- TTS Endpoint ---
+
+@pytest.mark.anyio
+async def test_create_speech_stream_success(client):
+    """
+    Tests the /speech endpoint to ensure it can successfully generate an audio stream.
+    """
+    test_client, mock_services = client
+    app = test_client.app  # Get the FastAPI app from the TestClient
+    
+    # Arrange: Define the text to convert and mock the service's response.
+    text_to_speak = "Hello, world!"
+    
+    # Define the async generator
+    async def mock_audio_generator():
+        yield b'chunk1'
+        yield b'chunk2'
+        yield b'chunk3'
+    
+    # Properly mock the method to return the generator
+    mock_services.tts_service.create_speech_stream = lambda text: mock_audio_generator()
+    
+    # Use AsyncClient with ASGITransport to send request to the FastAPI app
+    async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as ac:
+        response = await ac.post("/speech", json={"text": text_to_speak})
+
+    # Assert: Check status code and content
+    assert response.status_code == 200
+    assert response.headers["content-type"] == "audio/wav"
+    assert response.content == b"chunk1chunk2chunk3"
\ No newline at end of file

diff --git a/ai-hub/app/api/dependencies.py b/ai-hub/app/api/dependencies.py
index e1f2be7..e1aecde 100644
--- a/ai-hub/app/api/dependencies.py
+++ b/ai-hub/app/api/dependencies.py
@@ -6,8 +6,10 @@
 from app.core.retrievers.base_retriever import Retriever
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService
 from app.core.vector_store.faiss_store import FaissVectorStore
 
+
 # This is a dependency
 def get_db():
     db = SessionLocal()
@@ -25,9 +27,10 @@
 
 
 class ServiceContainer:
-    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever]):
+    def __init__(self, vector_store: FaissVectorStore, retrievers: List[Retriever], tts_service: TTSService):
         # Initialize all services within the container
         self.document_service = DocumentService(vector_store=vector_store)
         self.rag_service = RAGService(
             retrievers=retrievers
-        )
\ No newline at end of file
+        )
+        self.tts_service = tts_service
diff --git a/ai-hub/app/api/routes.py b/ai-hub/app/api/routes.py
index 6d216ec..40860aa 100644
--- a/ai-hub/app/api/routes.py
+++ b/ai-hub/app/api/routes.py
@@ -1,8 +1,11 @@
 from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
-from app.api.dependencies import ServiceContainer 
+from app.api.dependencies import ServiceContainer
 from app.api.dependencies import get_db
 from app.api import schemas
+from starlette.concurrency import run_in_threadpool
+
 
 def create_api_router(services: ServiceContainer) -> APIRouter:
     """
@@ -15,15 +18,12 @@
         return {"status": "AI Model Hub is running!"}
 
     # --- Session Management Endpoints ---
-    
+
     @router.post("/sessions", response_model=schemas.Session, summary="Create a New Chat Session", tags=["Sessions"])
     def create_session(
         request: schemas.SessionCreate,
         db: Session = Depends(get_db)
     ):
-        """
-        Starts a new conversation session and returns its details.
-        """
         try:
             new_session = services.rag_service.create_session(
                 db=db, 
@@ -34,40 +34,29 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"Failed to create session: {e}")
 
-    # --- Session Management Endpoints ---
     @router.post("/sessions/{session_id}/chat", response_model=schemas.ChatResponse, summary="Send a Message in a Session", tags=["Sessions"])
     async def chat_in_session(
         session_id: int,
         request: schemas.ChatRequest,
         db: Session = Depends(get_db)
     ):
-        """
-        Sends a message within an existing session and gets a contextual response.
-        
-        The 'model' and 'load_faiss_retriever' can now be specified in the request body.
-        """
         try:
             response_text, model_used = await services.rag_service.chat_with_rag(
                 db=db,
                 session_id=session_id,
                 prompt=request.prompt,
                 model=request.model,
-                load_faiss_retriever=request.load_faiss_retriever # Pass the new parameter to the service
+                load_faiss_retriever=request.load_faiss_retriever
             )
             return schemas.ChatResponse(answer=response_text, model_used=model_used)
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred during chat: {e}")
 
-
     @router.get("/sessions/{session_id}/messages", response_model=schemas.MessageHistoryResponse, summary="Get Session Chat History", tags=["Sessions"])
     def get_session_messages(session_id: int, db: Session = Depends(get_db)):
-        """
-        Retrieves the full message history for a specific session.
-        """
         try:
-            # Note: You'll need to add a `get_message_history` method to your RAGService.
             messages = services.rag_service.get_message_history(db=db, session_id=session_id)
-            if messages is None: # Service can return None if session not found
+            if messages is None:
                 raise HTTPException(status_code=404, detail=f"Session with ID {session_id} not found.")
             
             return schemas.MessageHistoryResponse(session_id=session_id, messages=messages)
@@ -77,7 +66,6 @@
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
     # --- Document Management Endpoints ---
-    # (These endpoints remain unchanged)
     
     @router.post("/documents", response_model=schemas.DocumentResponse, summary="Add a New Document", tags=["Documents"])
     def add_document(doc: schemas.DocumentCreate, db: Session = Depends(get_db)):
@@ -114,4 +102,26 @@
         except Exception as e:
             raise HTTPException(status_code=500, detail=f"An error occurred: {e}")
 
-    return router
+    # --- TTS Endpoint ---
+    @router.post(
+        "/speech",
+        summary="Generate a speech stream from text",
+        tags=["TTS"],
+        response_description="A stream of audio bytes in WAV format",
+    )
+    async def create_speech_stream(request: schemas.SpeechRequest):
+        """
+        Generates an audio stream from the provided text using the TTS service.
+        """
+        try:
+            # Use run_in_threadpool to turn the synchronous generator into an
+            # async generator that StreamingResponse can handle.
+            audio_stream = await run_in_threadpool(
+                services.tts_service.create_speech_stream, text=request.text
+            )
+            return StreamingResponse(audio_stream, media_type="audio/wav")
+        except Exception as e:
+            raise HTTPException(
+                status_code=500, detail=f"Failed to generate speech: {e}"
+            )
+    return router
\ No newline at end of file
diff --git a/ai-hub/app/api/schemas.py b/ai-hub/app/api/schemas.py
index 419aa44..31fb342 100644
--- a/ai-hub/app/api/schemas.py
+++ b/ai-hub/app/api/schemas.py
@@ -73,4 +73,7 @@
 class MessageHistoryResponse(BaseModel):
     """Defines the response for retrieving a session's chat history."""
     session_id: int
-    messages: List[Message]
\ No newline at end of file
+    messages: List[Message]
+
+class SpeechRequest(BaseModel):
+    text: str
\ No newline at end of file
diff --git a/ai-hub/app/app.py b/ai-hub/app/app.py
index a747ae6..a56fba6 100644
--- a/ai-hub/app/app.py
+++ b/ai-hub/app/app.py
@@ -6,12 +6,14 @@
 from app.config import settings
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.vector_store.embedder.factory import get_embedder_from_config
+from app.core.providers.factory import get_tts_provider
 from app.core.retrievers.faiss_db_retriever import FaissDBRetriever
 from app.core.retrievers.base_retriever import Retriever
 from app.db.session import create_db_and_tables
 from app.api.routes import create_api_router
 from app.utils import print_config 
 from app.api.dependencies import ServiceContainer
+from app.core.services.tts import TTSService
 # Note: The llm_clients import and initialization are removed as they
 # are not used in RAGService's constructor based on your services.py
 # from app.core.llm_clients import DeepSeekClient, GeminiClient
@@ -64,14 +66,30 @@
         
     )
     
+    # CRITICAL FIX: Assign the vector_store to the app state so it can be saved on shutdown.
+    app.state.vector_store = vector_store
+    
     # 3. Create the FaissDBRetriever, regardless of the embedder type
     retrievers: List[Retriever] = [
         FaissDBRetriever(vector_store=vector_store),
     ]
     
-    # 4. Initialize the Service Container
-
-    services = ServiceContainer(vector_store, retrievers)
+    # --- New TTS Initialization ---
+    # 4. Get the concrete TTS provider from the factory
+    tts_provider = get_tts_provider(
+        provider_name=settings.TTS_PROVIDER,
+        api_key=settings.TTS_API_KEY
+    )
+    
+    # 5. Initialize the TTSService
+    tts_service = TTSService(tts_provider=tts_provider)
+    
+    # 6. Initialize the Service Container with all services
+    services = ServiceContainer(
+        vector_store=vector_store, 
+        retrievers=retrievers,
+        tts_service=tts_service  # Pass the new TTS service instance
+    )
 
     # Create and include the API router, injecting the service
     api_router = create_api_router(services=services)
diff --git a/ai-hub/app/core/llm_providers.py b/ai-hub/app/core/llm_providers.py
deleted file mode 100644
index 09364b3..0000000
--- a/ai-hub/app/core/llm_providers.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# import httpx
-# import logging
-# import json
-# from abc import ABC, abstractmethod
-# from openai import OpenAI
-# from typing import final
-# from app.config import settings # <-- Import the centralized settings
-
-# # --- 1. Initialize API Clients from Central Config ---
-# # All environment variable access is now gone from this file.
-# deepseek_client = OpenAI(api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
-# GEMINI_URL = f"https://generativelanguage.googleapis.com/v1beta/models/{settings.GEMINI_MODEL_NAME}:generateContent?key={settings.GEMINI_API_KEY}"
-
-
-# # --- 2. Provider Interface and Implementations (Unchanged) ---
-# class LLMProvider(ABC):
-#     """Abstract base class ('Interface') for all LLM providers."""
-#     @abstractmethod
-#     async def generate_response(self, prompt: str) -> str:
-#         """Generates a response from the LLM."""
-#         pass
-
-# @final
-# class DeepSeekProvider(LLMProvider):
-#     """Provider for the DeepSeek API."""
-#     def __init__(self, model_name: str):
-#         self.model = model_name
-
-#     async def generate_response(self, prompt: str) -> str:
-#         messages = [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt}]
-#         try:
-#             chat_completion = deepseek_client.chat.completions.create(model=self.model, messages=messages)
-#             return chat_completion.choices[0].message.content
-#         except Exception as e:
-#             logging.error("DeepSeek Provider Error", exc_info=True)
-#             raise
-
-# @final
-# class GeminiProvider(LLMProvider):
-#     """Provider for the Google Gemini API."""
-#     def __init__(self, api_url: str):
-#         self.url = api_url
-
-#     async def generate_response(self, prompt: str) -> str:
-#         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-#         headers = {"Content-Type": "application/json"}
-#         try:
-#             async with httpx.AsyncClient() as client:
-#                 response = await client.post(self.url, json=payload, headers=headers)
-#                 response.raise_for_status()
-#                 data = response.json()
-#                 return data['candidates'][0]['content']['parts'][0]['text']
-#         except Exception as e:
-#             logging.error("Gemini Provider Error", exc_info=True)
-#             raise
-
-# # --- 3. The Factory Function ---
-# # The dictionary of providers is now built using values from the settings object.
-# _providers = {
-#     "deepseek": DeepSeekProvider(model_name=settings.DEEPSEEK_MODEL_NAME),
-#     "gemini": GeminiProvider(api_url=GEMINI_URL)
-# }
-
-# def get_llm_provider(model_name: str) -> LLMProvider:
-#     """Factory function to get the appropriate, pre-configured LLM provider."""
-#     provider = _providers.get(model_name)
-#     if not provider:
-#         raise ValueError(f"Unsupported model provider: '{model_name}'. Supported providers are: {list(_providers.keys())}")
-#     return provider
\ No newline at end of file
diff --git a/ai-hub/app/core/providers/factory.py b/ai-hub/app/core/providers/factory.py
index 7cf5119..53f3651 100644
--- a/ai-hub/app/core/providers/factory.py
+++ b/ai-hub/app/core/providers/factory.py
@@ -2,6 +2,7 @@
 from .base import LLMProvider,TTSProvider
 from .llm.deepseek import DeepSeekProvider
 from .llm.gemini import GeminiProvider
+from .tts.gemini import GeminiTTSProvider
 from openai import AsyncOpenAI
 
 # --- 1. Initialize API Clients from Central Config ---
@@ -23,7 +24,7 @@
     return provider
 
 def get_tts_provider(provider_name: str, api_key: str) -> TTSProvider:
-    if provider_name == "gemini":
+    if provider_name == "google_genai":
         return GeminiTTSProvider(api_key=api_key)
     # Add other TTS providers here if needed
     raise ValueError(f"Unknown TTS provider: {provider_name}")
\ No newline at end of file
diff --git a/ai-hub/app/core/services/tts.py b/ai-hub/app/core/services/tts.py
new file mode 100644
index 0000000..c196795
--- /dev/null
+++ b/ai-hub/app/core/services/tts.py
@@ -0,0 +1,25 @@
+from typing import AsyncGenerator
+from app.core.providers.base import TTSProvider
+
+class TTSService:
+    """
+    Service class for generating speech from text using a TTS provider.
+    """
+    def __init__(self, tts_provider: TTSProvider):
+        """
+        Initializes the TTSService with a concrete TTS provider.
+        """
+        self.tts_provider = tts_provider
+
+    async def create_speech_stream(self, text: str) -> AsyncGenerator[bytes, None]:
+        """
+        Generates a stream of audio bytes from the given text using the configured
+        TTS provider.
+
+        Args:
+            text: The text to be converted to speech.
+
+        Returns:
+            An async generator that yields chunks of audio bytes.
+        """
+        return self.tts_provider.generate_speech(text)
\ No newline at end of file
diff --git a/ai-hub/integration_tests/conftest.py b/ai-hub/integration_tests/conftest.py
new file mode 100644
index 0000000..5d51010
--- /dev/null
+++ b/ai-hub/integration_tests/conftest.py
@@ -0,0 +1,57 @@
+import httpx
+import pytest_asyncio
+
+BASE_URL = "http://127.0.0.1:8000"
+
+@pytest_asyncio.fixture(scope="session")
+def base_url():
+    """Fixture to provide the base URL for the tests."""
+    return BASE_URL
+
+@pytest_asyncio.fixture(scope="function") # <-- Change scope to "function"
+async def http_client():
+    """
+    Fixture to provide an async HTTP client for all tests in the session.
+    A new client is created and closed properly using a try/finally block
+    to prevent "Event loop is closed" errors.
+    """
+    client = httpx.AsyncClient(base_url=BASE_URL, timeout=60.0)
+    try:
+        yield client
+    finally:
+        await client.aclose()
+
+@pytest_asyncio.fixture(scope="function")
+async def session_id(http_client):
+    """
+    Creates a new session before a test and cleans it up after.
+    Returns the session ID.
+    """
+    payload = {"user_id": "integration_tester", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    yield session_id
+    # No explicit session deletion is needed for this example,
+    # as sessions are typically managed by a database lifecycle.
+
+@pytest_asyncio.fixture(scope="function")
+async def document_id(http_client):
+    """
+    Creates a new document before a test and ensures it's deleted afterward.
+    Returns the document ID.
+    """
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    response = await http_client.post("/documents", json=doc_data)
+    assert response.status_code == 200
+    try:
+        message = response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+        
+    yield document_id
+    
+    # Teardown: Delete the document after the test
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_documents.py b/ai-hub/integration_tests/test_documents.py
new file mode 100644
index 0000000..c6d4362
--- /dev/null
+++ b/ai-hub/integration_tests/test_documents.py
@@ -0,0 +1,33 @@
+import pytest
+
+@pytest.mark.asyncio
+async def test_document_lifecycle(http_client):
+    """
+    Tests the full lifecycle of a document: add, list, and delete.
+    This is run as a single, sequential test for a clean state.
+    """
+    print("\n--- Running test_document_lifecycle ---")
+    
+    # 1. Add a new document
+    doc_data = {"title": "Lifecycle Test Doc", "text": "This doc will be listed and deleted."}
+    add_response = await http_client.post("/documents", json=doc_data)
+    assert add_response.status_code == 200
+    try:
+        message = add_response.json().get("message", "")
+        document_id = int(message.split(" with ID ")[-1])
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    print(f"✅ Document for lifecycle test created with ID: {document_id}")
+    
+    # 2. List all documents and check if the new document is present
+    list_response = await http_client.get("/documents")
+    assert list_response.status_code == 200
+    ids_in_response = {doc["id"] for doc in list_response.json()["documents"]}
+    assert document_id in ids_in_response
+    print("✅ Document list test passed.")
+    
+    # 3. Delete the document
+    delete_response = await http_client.delete(f"/documents/{document_id}")
+    assert delete_response.status_code == 200
+    assert delete_response.json()["document_id"] == document_id
+    print("✅ Document delete test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_misc.py b/ai-hub/integration_tests/test_misc.py
new file mode 100644
index 0000000..3ff7925
--- /dev/null
+++ b/ai-hub/integration_tests/test_misc.py
@@ -0,0 +1,36 @@
+import pytest
+import httpx
+
+@pytest.mark.asyncio
+async def test_root_endpoint(http_client):
+    """
+    Tests if the root endpoint is alive and returns the correct status message.
+    """
+    print("\n--- Running test_root_endpoint ---")
+    response = await http_client.get("/")
+    assert response.status_code == 200
+    assert response.json() == {"status": "AI Model Hub is running!"}
+    print("✅ Root endpoint test passed.")
+
+# @pytest.mark.asyncio
+# async def test_create_speech_stream(http_client):
+#     """
+#     Tests the /speech endpoint for a successful audio stream response.
+#     """
+#     print("\n--- Running test_create_speech_stream ---")
+#     url = "/speech"
+#     payload = {"text": "Hello, world!"}
+
+#     # The `stream=True` parameter tells httpx to not read the entire response body
+#     # at once. We'll handle it manually to check for content.
+#     async with http_client.stream("POST", url, json=payload) as response:
+#         assert response.status_code == 200, f"Speech stream request failed. Response: {response.text}"
+#         assert response.headers.get("content-type") == "audio/wav"
+
+#         # Check that the response body is not empty by iterating over chunks.
+#         content_length = 0
+#         async for chunk in response.aiter_bytes():
+#             content_length += len(chunk)
+            
+#         assert content_length > 0
+#     print("✅ TTS stream test passed.")
\ No newline at end of file
diff --git a/ai-hub/integration_tests/test_sessions.py b/ai-hub/integration_tests/test_sessions.py
new file mode 100644
index 0000000..44d63d7
--- /dev/null
+++ b/ai-hub/integration_tests/test_sessions.py
@@ -0,0 +1,107 @@
+import pytest
+
+# Test prompts and data
+CONTEXT_PROMPT = "Who is the CEO of Microsoft?"
+FOLLOW_UP_PROMPT = "When was he born?"
+RAG_DOC_TITLE = "Fictional Company History"
+RAG_DOC_TEXT = "The company AlphaCorp was founded in 2021 by Jane Doe. Their primary product is a smart home device called 'Nexus'."
+RAG_PROMPT = "Who founded AlphaCorp and what is their main product?"
+
+@pytest.mark.asyncio
+async def test_chat_in_session_lifecycle(http_client):
+    """
+    Tests a full session lifecycle from creation to conversational memory.
+    This test is a single, sequential unit.
+    """
+    print("\n--- Running test_chat_in_session_lifecycle ---")
+    
+    # 1. Create a new session
+    payload = {"user_id": "integration_tester_lifecycle", "model": "deepseek"}
+    response = await http_client.post("/sessions", json=payload)
+    assert response.status_code == 200
+    session_id = response.json()["id"]
+    print(f"✅ Session created successfully with ID: {session_id}")
+    
+    # 2. First chat turn to establish context
+    chat_payload_1 = {"prompt": CONTEXT_PROMPT}
+    response_1 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_1)
+    assert response_1.status_code == 200
+    assert "Satya Nadella" in response_1.json()["answer"]
+    assert response_1.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 1 (context) test passed.")
+    
+    # 3. Second chat turn (follow-up) to test conversational memory
+    chat_payload_2 = {"prompt": FOLLOW_UP_PROMPT}
+    response_2 = await http_client.post(f"/sessions/{session_id}/chat", json=chat_payload_2)
+    assert response_2.status_code == 200
+    assert "1967" in response_2.json()["answer"]
+    assert response_2.json()["model_used"] == "deepseek"
+    print("✅ Chat Turn 2 (follow-up) test passed.")
+    
+    # 4. Cleanup (optional, but good practice if not using a test database that resets)
+    # The session data would typically be cleaned up by the database teardown.
+
+@pytest.mark.asyncio
+async def test_chat_with_model_switch(http_client, session_id):
+    """Tests switching models within an existing session."""
+    print("\n--- Running test_chat_with_model_switch ---")
+    
+    # Send a message to the new session with a different model
+    payload_gemini = {"prompt": "What is the capital of France?", "model": "gemini"}
+    response_gemini = await http_client.post(f"/sessions/{session_id}/chat", json=payload_gemini)
+    assert response_gemini.status_code == 200
+    assert "Paris" in response_gemini.json()["answer"]
+    assert response_gemini.json()["model_used"] == "gemini"
+    print("✅ Chat (Model Switch to Gemini) test passed.")
+
+    # Switch back to the original model
+    payload_deepseek = {"prompt": "What is the largest ocean?", "model": "deepseek"}
+    response_deepseek = await http_client.post(f"/sessions/{session_id}/chat", json=payload_deepseek)
+    assert response_deepseek.status_code == 200
+    assert "Pacific Ocean" in response_deepseek.json()["answer"]
+    assert response_deepseek.json()["model_used"] == "deepseek"
+    print("✅ Chat (Model Switch back to DeepSeek) test passed.")
+
+@pytest.mark.asyncio
+async def test_chat_with_document_retrieval(http_client):
+    """
+    Tests injecting a document and using it for retrieval-augmented generation.
+    This test creates its own session and document for isolation.
+    """
+    print("\n--- Running test_chat_with_document_retrieval ---")
+    
+    # Create a new session for this RAG test
+    session_response = await http_client.post("/sessions", json={"user_id": "rag_tester", "model": "deepseek"})
+    assert session_response.status_code == 200
+    rag_session_id = session_response.json()["id"]
+    
+    # Add a new document with specific content for retrieval
+    doc_data = {"title": RAG_DOC_TITLE, "text": RAG_DOC_TEXT}
+    add_doc_response = await http_client.post("/documents", json=doc_data)
+    assert add_doc_response.status_code == 200
+    try:
+        message = add_doc_response.json().get("message", "")
+        rag_document_id = int(message.split(" with ID ")[-1])
+        print(f"Document for RAG created with ID: {rag_document_id}")
+    except (ValueError, IndexError):
+        pytest.fail("Could not parse document ID from response message.")
+    
+    try:
+        chat_payload = {
+            "prompt": RAG_PROMPT,
+            "document_id": rag_document_id,
+            "model": "deepseek",
+            "load_faiss_retriever": True
+        }
+        chat_response = await http_client.post(f"/sessions/{rag_session_id}/chat", json=chat_payload)
+        
+        assert chat_response.status_code == 200
+        chat_data = chat_response.json()
+        assert "Jane Doe" in chat_data["answer"]
+        assert "Nexus" in chat_data["answer"]
+        print("✅ Chat with document retrieval test passed.")
+    finally:
+        # Clean up the document after the test
+        delete_response = await http_client.delete(f"/documents/{rag_document_id}")
+        assert delete_response.status_code == 200
+        print(f"Document {rag_document_id} deleted successfully.")
\ No newline at end of file
diff --git a/ai-hub/run_integration_tests.sh b/ai-hub/run_integration_tests.sh
index 65062f6..ef45dc0 100644
--- a/ai-hub/run_integration_tests.sh
+++ b/ai-hub/run_integration_tests.sh
@@ -4,15 +4,45 @@
 # It starts the FastAPI server, runs the specified tests, and then shuts down the server.
 
 # --- Configuration ---
-# Set the default path for tests to run. This will be used if no argument is provided.
-DEFAULT_TEST_PATH="integration_tests/"
-# You can override the default with a command-line argument, e.g., './run_integration_tests.sh tests/test_app.py'
-TEST_PATH=${1:-$DEFAULT_TEST_PATH}
+# You can define aliases for your test file paths here.
+TEST_SUITES=(
+  "All tests"
+  "integration_tests/test_sessions.py"
+  "integration_tests/test_documents.py"
+  "integration_tests/test_misc.py"
+)
+TEST_PATHS=(
+  "integration_tests/"
+  "integration_tests/test_sessions.py"
+  "integration_tests/test_documents.py"
+  "integration_tests/test_misc.py"
+)
 
 export DB_MODE=sqlite
 export LOCAL_DB_PATH="data/integration_test_ai_hub.db"
 export FAISS_INDEX_PATH="data/integration_test_faiss_index.bin"
 
+# --- User Interaction ---
+echo "--- AI Hub Test Runner ---"
+echo "Select a test suite to run:"
+
+# Display the menu options
+for i in "${!TEST_SUITES[@]}"; do
+  echo "$((i+1)). ${TEST_SUITES[$i]}"
+done
+
+# Prompt for user input
+read -p "Enter the number of your choice (1-${#TEST_SUITES[@]}): " choice
+
+# Validate input and set the TEST_PATH
+if [[ "$choice" =~ ^[0-9]+$ ]] && [ "$choice" -ge 1 ] && [ "$choice" -le "${#TEST_SUITES[@]}" ]; then
+  TEST_PATH=${TEST_PATHS[$((choice-1))]}
+  echo "You have selected: ${TEST_SUITES[$((choice-1))]}"
+else
+  echo "Invalid choice. Running all tests by default."
+  TEST_PATH=${TEST_PATHS[0]}
+fi
+
 # --- Pre-test Cleanup ---
 # Check for and remove old test files to ensure a clean test environment.
 echo "--- Checking for and removing old test files ---"
@@ -62,4 +92,4 @@
 
 # The 'trap' will automatically call the cleanup function now.
 # Exit with the same code as the test script (0 for success, non-zero for failure).
-exit $TEST_EXIT_CODE
+exit $TEST_EXIT_CODE
\ No newline at end of file
diff --git a/ai-hub/tests/api/test_dependencies.py b/ai-hub/tests/api/test_dependencies.py
index 65490e8..d2e0472 100644
--- a/ai-hub/tests/api/test_dependencies.py
+++ b/ai-hub/tests/api/test_dependencies.py
@@ -9,6 +9,7 @@
 from app.api.dependencies import get_db, get_current_user, ServiceContainer
 from app.core.services.document import DocumentService
 from app.core.services.rag import RAGService
+from app.core.services.tts import TTSService # Added this import
 from app.core.vector_store.faiss_store import FaissVectorStore
 from app.core.retrievers.base_retriever import Retriever
 
@@ -97,11 +98,13 @@
     # The DocumentService constructor needs a .embedder attribute on the vector_store
     mock_vector_store.embedder = MagicMock()
     mock_retrievers = [MagicMock(spec=Retriever), MagicMock(spec=Retriever)]
+    mock_tts_service = MagicMock(spec=TTSService)
 
     # Act: Instantiate the ServiceContainer
     container = ServiceContainer(
         vector_store=mock_vector_store,
-        retrievers=mock_retrievers
+        retrievers=mock_retrievers,
+        tts_service=mock_tts_service # Passing the mock TTS service
     )
 
     # Assert: Check if the services were created and configured correctly
@@ -110,3 +113,7 @@
     
     assert isinstance(container.rag_service, RAGService)
     assert container.rag_service.retrievers == mock_retrievers
+    
+    # Assert for the tts_service as well
+    assert isinstance(container.tts_service, TTSService)
+    assert container.tts_service == mock_tts_service
diff --git a/ai-hub/tests/api/test_routes.py b/ai-hub/tests/api/test_routes.py
index 8d841a2..6d5cd3e 100644
--- a/ai-hub/tests/api/test_routes.py
+++ b/ai-hub/tests/api/test_routes.py
@@ -1,15 +1,18 @@
-# tests/app/api/test_routes.py
+# tests/api/test_routes.py
 import pytest
 from unittest.mock import MagicMock, AsyncMock
 from fastapi import FastAPI
 from fastapi.testclient import TestClient
 from sqlalchemy.orm import Session
 from datetime import datetime
+from httpx import AsyncClient, ASGITransport
+
 
 # Import the dependencies and router factory
 from app.api.dependencies import get_db, ServiceContainer
 from app.core.services.rag import RAGService
 from app.core.services.document import DocumentService
+from app.core.services.tts import TTSService 
 from app.api.routes import create_api_router
 from app.db import models # Import your SQLAlchemy models
 
@@ -24,11 +27,15 @@
     # Mock individual services
     mock_rag_service = MagicMock(spec=RAGService)
     mock_document_service = MagicMock(spec=DocumentService)
-
+    
+    # Use AsyncMock for the TTS service since its methods are async
+    mock_tts_service = MagicMock(spec=TTSService)
+    
     # Create a mock ServiceContainer that holds the mocked services
     mock_services = MagicMock(spec=ServiceContainer)
     mock_services.rag_service = mock_rag_service
     mock_services.document_service = mock_document_service
+    mock_services.tts_service = mock_tts_service
     
     # Mock the database session
     mock_db_session = MagicMock(spec=Session)
@@ -80,8 +87,6 @@
     assert response.status_code == 200
     assert response.json() == {"answer": "Mocked response", "model_used": "deepseek"}
     
-    # Verify that chat_with_rag was called with the default model 'deepseek'
-    # and the default load_faiss_retriever=False
     mock_services.rag_service.chat_with_rag.assert_called_once_with(
         db=mock_services.rag_service.chat_with_rag.call_args.kwargs['db'],
         session_id=42,
@@ -102,13 +107,12 @@
     assert response.status_code == 200
     assert response.json() == {"answer": "Mocked response from Gemini", "model_used": "gemini"}
     
-    # Verify that chat_with_rag was called with the specified model 'gemini'
     mock_services.rag_service.chat_with_rag.assert_called_once_with(
         db=mock_services.rag_service.chat_with_rag.call_args.kwargs['db'],
         session_id=42,
         prompt="Hello there, Gemini!",
         model="gemini",
-        load_faiss_retriever=False # It should still default to False
+        load_faiss_retriever=False
     )
 
 def test_chat_in_session_with_faiss_retriever(client):
@@ -126,29 +130,25 @@
     assert response.status_code == 200
     assert response.json() == {"answer": "Response with context", "model_used": "deepseek"}
     
-    # Verify that chat_with_rag was called with the correct parameters
     mock_services.rag_service.chat_with_rag.assert_called_once_with(
         db=mock_services.rag_service.chat_with_rag.call_args.kwargs['db'],
         session_id=42,
         prompt="What is RAG?",
-        model="deepseek", # The model still defaults to deepseek
-        load_faiss_retriever=True # Verify that the retriever was explicitly enabled
+        model="deepseek",
+        load_faiss_retriever=True
     )
 
 def test_get_session_messages_success(client):
     """Tests retrieving the message history for a session."""
     test_client, mock_services = client
-    # Arrange: Mock the service to return a list of message objects
     mock_history = [
         models.Message(sender="user", content="Hello", created_at=datetime.now()),
         models.Message(sender="assistant", content="Hi there!", created_at=datetime.now())
     ]
     mock_services.rag_service.get_message_history.return_value = mock_history
     
-    # Act
     response = test_client.get("/sessions/123/messages")
     
-    # Assert
     assert response.status_code == 200
     response_data = response.json()
     assert response_data["session_id"] == 123
@@ -163,13 +163,10 @@
 def test_get_session_messages_not_found(client):
     """Tests retrieving messages for a session that does not exist."""
     test_client, mock_services = client
-    # Arrange: Mock the service to return None, indicating the session wasn't found
     mock_services.rag_service.get_message_history.return_value = None
     
-    # Act
     response = test_client.get("/sessions/999/messages")
     
-    # Assert
     assert response.status_code == 404
     assert response.json()["detail"] == "Session with ID 999 not found."
 
@@ -210,3 +207,34 @@
     mock_services.document_service.delete_document.return_value = None
     response = test_client.delete("/documents/999")
     assert response.status_code == 404
+    
+# --- TTS Endpoint ---
+
+@pytest.mark.anyio
+async def test_create_speech_stream_success(client):
+    """
+    Tests the /speech endpoint to ensure it can successfully generate an audio stream.
+    """
+    test_client, mock_services = client
+    app = test_client.app  # Get the FastAPI app from the TestClient
+    
+    # Arrange: Define the text to convert and mock the service's response.
+    text_to_speak = "Hello, world!"
+    
+    # Define the async generator
+    async def mock_audio_generator():
+        yield b'chunk1'
+        yield b'chunk2'
+        yield b'chunk3'
+    
+    # Properly mock the method to return the generator
+    mock_services.tts_service.create_speech_stream = lambda text: mock_audio_generator()
+    
+    # Use AsyncClient with ASGITransport to send request to the FastAPI app
+    async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as ac:
+        response = await ac.post("/speech", json={"text": text_to_speak})
+
+    # Assert: Check status code and content
+    assert response.status_code == 200
+    assert response.headers["content-type"] == "audio/wav"
+    assert response.content == b"chunk1chunk2chunk3"
\ No newline at end of file
diff --git a/ai-hub/tests/test_app.py b/ai-hub/tests/test_app.py
index 73eb307..c0bb4b4 100644
--- a/ai-hub/tests/test_app.py
+++ b/ai-hub/tests/test_app.py
@@ -12,6 +12,7 @@
 from app.api.dependencies import get_db, ServiceContainer
 from app.db import models
 from app.core.retrievers.base_retriever import Retriever
+from app.core.vector_store.faiss_store import FaissVectorStore
 
 # Define a constant for the dimension to ensure consistency
 TEST_DIMENSION = 768
@@ -30,11 +31,10 @@
 # We patch ServiceContainer directly to control its instantiation in create_app
 @patch('app.app.ServiceContainer')
 @patch('app.app.create_db_and_tables')
-@patch('app.app.FaissVectorStore.save_index')
 @patch('app.app.print_config')
 @patch('app.core.vector_store.embedder.factory.get_embedder_from_config')
 @patch('faiss.read_index') # This patch is for the FaissVectorStore initialization
-def test_read_root(mock_read_index, mock_get_embedder, mock_print_config, mock_save_index, mock_create_db, mock_service_container):
+def test_read_root(mock_read_index, mock_get_embedder, mock_print_config, mock_create_db, mock_service_container):
     """Test the root endpoint to ensure it's running."""
     # Arrange: We patch the embedder and faiss calls to prevent real logic
     mock_read_index.return_value = MagicMock()
@@ -53,8 +53,6 @@
     assert response.status_code == 200
     assert response.json() == {"status": "AI Model Hub is running!"}
 
-
-# We patch ServiceContainer directly to control its instantiation in create_app
 @patch('app.app.ServiceContainer')
 @patch('app.app.create_db_and_tables')
 @patch('app.core.vector_store.embedder.factory.get_embedder_from_config')
@@ -334,3 +332,27 @@
     assert response.status_code == 404
     assert response.json()["detail"] == "Document with ID 999 not found."
     mock_services.document_service.delete_document.assert_called_once_with(db=mock_db, document_id=999)
+
+# FIX: Add a new test to explicitly check the application shutdown behavior
+@patch('app.core.vector_store.faiss_store.FaissVectorStore.save_index')
+@patch('app.app.ServiceContainer')
+@patch('app.app.create_db_and_tables')
+@patch('app.app.print_config')
+@patch('app.core.vector_store.embedder.factory.get_embedder_from_config')
+@patch('faiss.read_index')
+def test_shutdown_saves_index(mock_read_index, mock_get_embedder, mock_print_config, mock_create_db, mock_service_container, mock_save_index):
+    """
+    Tests that the FAISS index is saved on application shutdown.
+    """
+    # Arrange
+    mock_read_index.return_value = MagicMock()
+    mock_get_embedder.return_value = MagicMock()
+    
+    # Create the app and let the lifespan events run
+    app = create_app()
+    with TestClient(app) as client:
+        # Act: The lifespan shutdown event will run when the 'with' block is exited
+        pass
+    
+    # Assert
+    mock_save_index.assert_called_once()