Update app.py

app.py

@@ -27,8 +27,9 @@ logger = logging.getLogger(__name__)
 # Global variables
 rag_system: Optional[MultimodalRAGSystem] = None
 
-# Store last question-answer pair for simple follow-up
-last_qa_context: Optional[str] = None
+# Keep short conversation history 
+chat_history: List[Dict[str, str]] = []
+MAX_HISTORY_TURNS = 3
 
 # Lifecycle management
 @asynccontextmanager
@@ -124,23 +125,24 @@ async def health_check():
 
 @app.post("/query", response_model=QueryResponse, tags=["Query"])
 async def query_rag(request: QueryRequest):
-    global last_qa_context
+    global chat_history
 
     if not rag_system:
         raise HTTPException(status_code=503, detail="RAG system not initialized.")
 
     try:
-        # Pass ONLY the question to retrieval
         result = rag_system.ask(
-            request.question,
-            context=last_qa_context
+            question=request.question,
+            chat_history=chat_history
         )
 
-        # Save lightweight context for next turn
-        last_qa_context = (
-            f"Q: {request.question}\n"
-            f"A: {result['answer'][:500]}"  # truncate for safety
-        )
+        # Update history
+        chat_history.append({"role": "user", "content": request.question})
+        chat_history.append({"role": "assistant", "content": result["answer"]})
+
+        # Trim history safely
+        if len(chat_history) > MAX_HISTORY_TURNS * 2:
+            chat_history = chat_history[-MAX_HISTORY_TURNS * 2:]
 
         return QueryResponse(
             answer=result["answer"],
@@ -154,6 +156,7 @@ async def query_rag(request: QueryRequest):
         raise HTTPException(status_code=500, detail=str(e))
 
 
+
 if __name__ == "__main__":
     import uvicorn
     uvicorn.run(app, host="0.0.0.0", port=7860)