Update llm_handler.py

llm_handler.py

@@ -125,20 +125,29 @@ def get_rag_response(query: str, session_id: str = None) -> tuple[str, str]:
     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()
-    results = chroma_collection.query(
-        query_embeddings=[query_embedding],
-        n_results=3,
-    )
-    
-    # Get the 'metadatas' which contain the full internship details
-    retrieved_docs = results.get('metadatas', [[]])[0]
-    context = "\n".join([str(doc) for doc in retrieved_docs])
+    # Get chat history
+    chat_history = get_chat_history(session_id)
+    is_first_message = len(chat_history) == 0
+
+    # Only retrieve context for the first message or when explicitly needed
+    context = ""
+    if is_first_message or any(word in query.lower() for word in ['internship', 'job', 'opportunity', 'skill', 'apply', 'stipend', 'duration']):
+        # Retrieve relevant documents from ChromaDB
+        query_embedding = encoder.encode([query])[0].tolist()
+        results = chroma_collection.query(
+            query_embeddings=[query_embedding],
+            n_results=3,
+        )
+        retrieved_docs = results.get('metadatas', [[]])[0]
+        context = "\n".join([str(doc) for doc in retrieved_docs])
+        print(f"🔍 Retrieved context for query (length: {len(context)})")
 
-    # 2. Prepare the conversation with chat history
-    system_prompt = """
-You are a helpful and friendly assistant for the PM Internship Scheme. 
+    # Build the conversation messages
+    messages = []
+    
+    # Add system prompt only for first message or when context is needed
+    if is_first_message or context:
+        system_content = """You are a helpful and friendly assistant for the PM Internship Scheme. 
 Your role is to guide users about internship opportunities, skills required, and preparation tips.
 
 Rules:
@@ -147,34 +156,32 @@ Rules:
 - Keep answers clear, natural, and helpful — aim for short but complete responses (3–6 sentences). 
 - Use a friendly, encouraging tone while staying professional.
 - 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 + """
+- When user says "the first one", "that internship", "it", etc., refer back to what was mentioned in the conversation history."""
 
-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.".
-"""
+        if context:
+            system_content += f"\n\nAvailable internship context for this query:\n{context}"
 
-    # Build the conversation messages
-    messages = [{"role": "system", "content": system_prompt}]
+        system_content += "\n\nIf 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.'."
+        
+        messages.append({"role": "system", "content": system_content})
+        print(f"📝 Added system prompt (with context: {bool(context)})")
     
-    # Add chat history (this is the crucial part!)
-    chat_history = get_chat_history(session_id)
+    # Add chat history
     for msg in chat_history:
         messages.append(msg)
     
-    # Add current user query (without the context prefix this time)
+    # Add current user query
     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]}...")
+    print(f"🔍 Debug - Sending {len(messages)} messages to LLM (reduced from full context each time)")
+    for i, msg in enumerate(messages[-3:], len(messages)-3):  # Show only last 3 messages in debug
+        print(f"  {i}: {msg['role']}: {msg['content'][:80]}...")
 
     try:
         completion = openrouter_client.chat.completions.create(
             model="x-ai/grok-4-fast",
             messages=messages,
-            max_tokens=500,  # Limit response length
+            max_tokens=500,
             temperature=0.7,
         )