Update src/rag.py

src/rag.py

@@ -466,5 +466,20 @@ class AACAssistant:
         response = self.chain.invoke(
             {"question": user_query, "emotion_analysis": emotion_analysis}
         )
-
-        return response["answer"]
+        assistant_marker = "</s> <|assistant|>" 
+        
+        # Find the last occurrence of the marker
+        marker_position = raw_full_answer.rfind(assistant_marker)
+        
+        if marker_position != -1:
+            # Extract the text *after* the marker
+            actual_response = raw_full_answer[marker_position + len(assistant_marker):].strip()
+        else:
+            # Fallback if the marker is not found in the response.
+            # This might happen if the LLM's output is unexpected or if the prompt structure changed.
+            print(f"WARNING: Assistant marker '{assistant_marker}' not found in raw answer. Returning raw answer as fallback.")
+            actual_response = raw_full_answer.strip() # Or handle as an error
+
+        print(f"DEBUG: process_query - Extracted assistant response: '{actual_response}'")
+        return actual_response
+        # return response["answer"]