Update llm_handler.py

llm_handler.py

@@ -106,15 +106,24 @@ def get_rag_response(query: str, session_id: str = None) -> tuple[str, str]:
     if not all([encoder, chroma_collection, openrouter_client]):
         return "Chatbot is not ready. Models or clients are not loaded.", session_id or create_chat_session()
 
-    # Create a new session if none provided
-    if session_id is None:
+def get_rag_response(query: str, session_id: str = None) -> tuple[str, str]:
+    """Generates a response using Retrieval-Augmented Generation with chat memory."""
+    if not all([encoder, chroma_collection, openrouter_client]):
+        return "Chatbot is not ready. Models or clients are not loaded.", session_id or create_chat_session()
+
+    # Create a new session ONLY if none provided
+    if session_id is None or session_id == "":
         session_id = create_chat_session()
         print(f"🆕 Created new chat session: {session_id}")
+    else:
+        print(f"🔄 Using existing session: {session_id}")
     
     # Validate session exists, create if it doesn't
     if session_id not in chat_sessions:
         chat_sessions[session_id] = []
-        print(f"🔄 Session {session_id} not found, created new one")
+        print(f"⚠️  Session {session_id} not found in memory, creating new one")
+    else:
+        print(f"✅ Found existing session with {len(chat_sessions[session_id])} messages")
 
     # 1. Retrieve relevant documents from ChromaDB
     query_embedding = encoder.encode([query])[0].tolist()
@@ -137,7 +146,11 @@ Rules:
 - If asked for such info, politely refuse and redirect them to the official PM Internship portal. 
 - Keep answers clear, natural, and helpful — aim for short but complete responses (3–6 sentences). 
 - Use a friendly, encouraging tone while staying professional.
-- Remember the conversation history and provide contextual responses.
+- IMPORTANT: Remember the conversation history and provide contextual responses based on what was discussed earlier.
+- When user says "the first one", "that internship", "it", etc., refer back to what was mentioned in the conversation history.
+
+Available internship context for this query:
+""" + context + """
 
 If the context doesn't have the answer, use your own general knowledge to provide a helpful response, even then if you are unable to answer the question, say: "I don't have that information, please check the official PM Internship portal.".
 """
@@ -145,13 +158,17 @@ If the context doesn't have the answer, use your own general knowledge to provid
     # Build the conversation messages
     messages = [{"role": "system", "content": system_prompt}]
     
-    # Add chat history
+    # Add chat history (this is the crucial part!)
     chat_history = get_chat_history(session_id)
-    messages.extend(chat_history)
+    for msg in chat_history:
+        messages.append(msg)
     
-    # Add current query with context
-    current_message = f"Context:\n{context}\n\nQuestion: {query}"
-    messages.append({"role": "user", "content": current_message})
+    # Add current user query (without the context prefix this time)
+    messages.append({"role": "user", "content": query})
+    
+    print(f"🔍 Debug - Messages being sent to LLM:")
+    for i, msg in enumerate(messages):
+        print(f"  {i}: {msg['role']}: {msg['content'][:100]}...")
 
     try:
         completion = openrouter_client.chat.completions.create(
@@ -163,10 +180,12 @@ If the context doesn't have the answer, use your own general knowledge to provid
         
         answer = completion.choices[0].message.content
         
-        # Add the conversation to chat history
+        # Add the conversation to chat history (store clean versions without context)
         add_to_chat_history(session_id, "user", query)
         add_to_chat_history(session_id, "assistant", answer)
         
+        print(f"💾 Added to history - Session {session_id} now has {len(chat_sessions[session_id])} messages")
+        
         return answer, session_id
         
     except Exception as e: