Update chatbot.py

chatbot.py

@@ -819,46 +819,43 @@ class RAGChatbot:
         return prompt
 
     def ask(self, question: str) -> str:
-        """Ask a question to the chatbot with learning from past conversations"""
+    """Ask a question to the chatbot with learning from past conversations"""
         if question.lower() in ["reset data", "reset"]:
             self.chat_history = []
             self.chat_embeddings = []
             self.chat_index = None
-            self.conversation_context = {'current_employee': None, 'last_mentioned_entities': []}  # ADD THIS LINE
+            self.conversation_context = {'current_employee': None, 'last_mentioned_entities': []}
             self._save_chat_history()
             return "Chat history has been reset."
-
-        # ADD THIS LINE:
+    
+        # Resolve pronouns before processing
         resolved_question = self._resolve_pronouns(question)
-
-        # CHANGE 'question' to 'resolved_question' in next line:
+        
+        # Extract query pattern for learning
         pattern = self._extract_query_pattern(resolved_question)
         self.query_patterns[pattern] += 1
-
-        # CHANGE 'question' to 'resolved_question':
-        relevant_past_chats = self._search_chat_history(resolved_question, k=10)
-
-        # CHANGE 'question' to 'resolved_question':
+    
+        # Search through past conversations for similar questions
+        relevant_past_chats = self._search_chat_history(resolved_question, k=5)
+    
+        # Retrieve relevant chunks (use resolved question for better retrieval)
         retrieved_data = self._retrieve(resolved_question, k=20)
-
-        # CHANGE 'question' to 'resolved_question':
+    
+        # Build prompt with both document context and learned information
         prompt = self._build_prompt(resolved_question, retrieved_data, relevant_past_chats)
-
-        # Generate response
-        messages = [{"role": "user", "content": prompt}]
-
-        response = self.llm_client.chat.completions.create(
+    
+        # ✅ CORRECT: Use text_generation for InferenceClient
+        answer = self.llm_client.text_generation(
+            prompt,
             model="meta-llama/Llama-3.1-8B-Instruct",
-            messages=messages,
-            max_tokens=512,
-            temperature=0.3
+            max_new_tokens=512,
+            temperature=0.3,
+            return_full_text=False
         )
-
-        answer = response.choices[0].message.content
-
-        # ADD THIS LINE:
+    
+        # Update conversation context after each exchange
         self._update_conversation_context(question, answer)
-
+    
         # Store in history with timestamp and metadata
         chat_entry = {
             'timestamp': datetime.now().isoformat(),
@@ -867,26 +864,26 @@ class RAGChatbot:
             'pattern': pattern,
             'used_past_context': len(relevant_past_chats) > 0
         }
-
+        
         self.chat_history.append(chat_entry)
-
+    
         # Update chat history index with new conversation
         new_text = f"Q: {question}\nA: {answer}"
         new_embedding = self.embeddings_model.encode([new_text])
-
+        
         if self.chat_index is None:
             dimension = new_embedding.shape[1]
             self.chat_index = faiss.IndexFlatL2(dimension)
             self.chat_embeddings = new_embedding
         else:
             self.chat_embeddings = np.vstack([self.chat_embeddings, new_embedding])
-
+        
         self.chat_index.add(np.array(new_embedding).astype('float32'))
-
+    
         # Save to disk after each conversation
         self._save_chat_history()
         self._save_learning_stats()
-
+    
         return answer
 
     def provide_feedback(self, question: str, rating: int):