Update app.py

app.py

@@ -69,10 +69,11 @@ class LLMGAIAAgent:
             # Clean up the response if needed
             response = self._clean_response(response)
             
-            return response
+            # Return JSON with final_answer key
+            return json.dumps({"final_answer": response})
         except Exception as e:
             print(f"Error generating response: {e}")
-            return self._fallback_response(question)
+            return json.dumps({"final_answer": self._fallback_response(question)})
     
     def _prepare_prompt(self, question: str) -> str:
         """Prepare an appropriate prompt based on the question type."""
@@ -160,7 +161,10 @@ class GAIAAgent:
         question_type = self._classify_question(question)
         
         # Use the appropriate handler
-        return self.handlers[question_type](question)
+        answer = self.handlers[question_type](question)
+        
+        # Return JSON with final_answer key
+        return json.dumps({"final_answer": answer})
     
     def _classify_question(self, question: str) -> str:
         """Classify the question into one of the supported types."""
@@ -340,9 +344,15 @@ class EvaluationRunner:
             try:
                 # Call agent with task_id parameter if supported
                 if hasattr(agent, '__code__') and 'task_id' in agent.__code__.co_varnames:
-                    submitted_answer = agent(question_text, task_id)
+                    json_response = agent(question_text, task_id)
                 else:
-                    submitted_answer = agent(question_text)
+                    json_response = agent(question_text)
+                
+                # Parse the JSON response
+                response_obj = json.loads(json_response)
+                
+                # Extract the final_answer for submission
+                submitted_answer = response_obj.get("final_answer", "")
                 
                 answers_payload.append({
                     "task_id": task_id, 
@@ -351,7 +361,8 @@ class EvaluationRunner:
                 results_log.append({
                     "Task ID": task_id, 
                     "Question": question_text, 
-                    "Submitted Answer": submitted_answer
+                    "Submitted Answer": submitted_answer,
+                    "Full Response": json_response
                 })
             except Exception as e:
                 print(f"Error running agent on task {task_id}: {e}")