Update app.py

app.py

@@ -1,7 +1,9 @@
 """
 app.py
 This script provides the Gradio web interface to run the evaluation.
-This version properly handles multimodal inputs including images, videos, and audio.
+## MODIFICATION: This version is simplified to work with the new agent architecture.
+It no longer performs file-type detection or prompt enhancement, as that responsibility
+has been moved into the agent's 'multimodal_router'.
 """
 
 import os
@@ -16,7 +18,7 @@ from agent import create_agent_executor
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
-# --- Helper function to parse the agent's output ---
+# --- Helper function to parse the agent's output (remains the same) ---
 def parse_final_answer(agent_response: str) -> str:
     match = re.search(r"FINAL ANSWER:\s*(.*)", agent_response, re.IGNORECASE | re.DOTALL)
     if match: return match.group(1).strip()
@@ -24,74 +26,12 @@ def parse_final_answer(agent_response: str) -> str:
     if lines: return lines[-1].strip()
     return "Could not parse a final answer."
 
-def detect_file_type(url: str) -> str:
-    """Detect the type of file from URL."""
-    if not url:
-        return "unknown"
-    
-    url_lower = url.lower()
-    
-    # Image extensions
-    if any(ext in url_lower for ext in ['.jpg', '.jpeg', '.png', '.gif', '.bmp', '.webp', '.svg']):
-        return "image"
-    
-    # Video extensions and YouTube
-    if any(domain in url_lower for domain in ['youtube.com', 'youtu.be', 'vimeo.com']):
-        return "youtube"
-    if any(ext in url_lower for ext in ['.mp4', '.avi', '.mov', '.wmv', '.flv', '.webm']):
-        return "video"
-    
-    # Audio extensions
-    if any(ext in url_lower for ext in ['.mp3', '.wav', '.flac', '.aac', '.ogg', '.m4a']):
-        return "audio"
-    
-    # Try to detect from headers if possible
-    try:
-        response = requests.head(url, timeout=5)
-        content_type = response.headers.get('content-type', '').lower()
-        
-        if 'image' in content_type:
-            return "image"
-        elif 'audio' in content_type:
-            return "audio"
-        elif 'video' in content_type:
-            return "video"
-    except:
-        pass
-    
-    return "unknown"
+## MODIFICATION: The `detect_file_type` function has been removed.
+## It is now redundant as this logic is handled inside the agent.
 
-def create_enhanced_prompt(question_text: str, file_url: str = None) -> str:
-    """Create an enhanced prompt that guides the agent to use appropriate tools."""
-    
-    if not file_url:
-        return question_text
-    
-    file_type = detect_file_type(file_url)
-    
-    if file_type == "image":
-        return f"""{question_text}
-
-[IMAGE ATTACHMENT]: {file_url}
-INSTRUCTION: There is an image attached to this question. You MUST use the 'describe_image' tool to analyze this image before answering the question."""
-
-    elif file_type == "youtube":
-        return f"""{question_text}
-
-[YOUTUBE VIDEO]: {file_url}
-INSTRUCTION: There is a YouTube video attached to this question. You MUST use the 'process_youtube_video' tool to analyze this video before answering the question."""
-
-    elif file_type == "audio":
-        return f"""{question_text}
-
-[AUDIO FILE]: {file_url}
-INSTRUCTION: There is an audio file attached to this question. You MUST use the 'process_audio_file' tool to analyze this audio before answering the question."""
-
-    else:
-        return f"""{question_text}
-
-[ATTACHMENT]: {file_url}
-INSTRUCTION: There is a file attachment. Analyze the URL and use the appropriate tool to process this content before answering the question."""
+## MODIFICATION: The `create_enhanced_prompt` function has been removed.
+## It was causing errors by trying to instruct the agent to use tools that no longer exist.
+## The agent is now responsible for handling the raw input itself.
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
     """
@@ -138,20 +78,21 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         
         print(f"\n--- Running Task {i+1}/{len(questions_data)} (ID: {task_id}) ---")
         
-        # Get file URL if it exists
         file_url = item.get("file_url")
         
-        # Create enhanced prompt that instructs the agent to use appropriate tools
-        full_question_text = create_enhanced_prompt(question_text, file_url)
-        
+        ## MODIFICATION: Prompt creation is now much simpler.
+        # We just combine the question and the URL into one string.
+        # The agent's multimodal_router will handle the rest.
         if file_url:
-            file_type = detect_file_type(file_url)
-            print(f"File detected: {file_url} (Type: {file_type})")
+            full_question_text = f"{question_text}\n\nHere is the relevant file: {file_url}"
+            print(f"File provided: {file_url}")
+        else:
+            full_question_text = question_text
         
-        print(f"Enhanced Prompt for Agent:\n{full_question_text}")
+        print(f"Raw Prompt for Agent:\n{full_question_text}")
 
         try:
-            # Pass the enhanced question to the agent
+            # Pass the simple, raw question to the agent
             result = agent_executor.invoke({"messages": [("user", full_question_text)]})
             
             raw_answer = result['messages'][-1].content
@@ -165,7 +106,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
                 "Task ID": task_id, 
                 "Question": question_text, 
                 "File URL": file_url or "None",
-                "File Type": detect_file_type(file_url) if file_url else "None",
                 "Submitted Answer": submitted_answer
             })
             
@@ -177,14 +117,13 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
                 "Task ID": task_id, 
                 "Question": question_text, 
                 "File URL": file_url or "None",
-                "File Type": detect_file_type(file_url) if file_url else "None",
                 "Submitted Answer": error_msg
             })
 
     if not answers_payload:
         return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
 
-    # 4. Prepare and 5. Submit
+    # 4. Prepare and 5. Submit (remains the same)
     submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
     print(f"\nSubmitting {len(answers_payload)} answers for user '{username}'...")
     try:
@@ -200,7 +139,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         print(status_message)
         return status_message, pd.DataFrame(results_log)
 
-# --- Gradio UI ---
+# --- Gradio UI (remains the same) ---
 with gr.Blocks(title="Multimodal Agent Evaluation") as demo:
     gr.Markdown("# Multimodal Agent Evaluation Runner")
     gr.Markdown("This agent can process images, YouTube videos, audio files, and perform web searches.")
@@ -212,10 +151,18 @@ with gr.Blocks(title="Multimodal Agent Evaluation") as demo:
         label="Questions and Agent Answers", 
         wrap=True, 
         row_count=10,
-        column_widths=[80, 200, 150, 80, 200]
+        # MODIFICATION: Removed the 'File Type' column as it's no longer detected here.
+        column_widths=[80, 250, 200, 250] 
     )
     
-    run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
+    # We also remove "File Type" from the results_log being displayed
+    def display_wrapper(profile):
+        status, df = run_and_submit_all(profile)
+        if df is not None and "File Type" in df.columns:
+            df = df.drop(columns=["File Type"])
+        return status, df
+
+    run_button.click(fn=display_wrapper, outputs=[status_output, results_table])
 
 if __name__ == "__main__":
     print("\n" + "-"*30 + " Multimodal App Starting " + "-"*30)