app.py

import os
import gradio as gr
import requests
import pandas as pd
from io import BytesIO
import re
import subprocess
import base64

# --- Tool-specific Imports ---
from pytube import YouTube
from langchain_huggingface import HuggingFaceInferenceAPI

# --- LangChain & Groq Imports ---
from groq import Groq
from langchain_groq import ChatGroq
from langchain.agents import AgentExecutor, create_tool_calling_agent
from langchain_tavily import TavilySearchResults
from langchain_core.prompts import ChatPromptTemplate
from langchain.tools import Tool

# --- Constants ---
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
TEMP_DIR = "/tmp"

# --- Tool Definition: Audio File Transcription ---
def transcribe_audio_file(task_id: str) -> str:
    # (This function is complete and correct from the previous version)
    print(f"Tool 'transcribe_audio_file' called with task_id: {task_id}")
    try:
        file_url = f"{DEFAULT_API_URL}/files/{task_id}"
        audio_response = requests.get(file_url)
        audio_response.raise_for_status()
        audio_bytes = BytesIO(audio_response.content)
        audio_bytes.name = f"{task_id}.mp3"
        client = Groq(api_key=os.getenv("GROQ_API_KEY"))
        transcription = client.audio.transcriptions.create(file=audio_bytes, model="whisper-large-v3", response_format="text")
        return str(transcription)
    except Exception as e:
        return f"Error during audio file transcription: {e}"

# --- Tool Definition: Video Transcription via FFmpeg ---
def transcribe_youtube_video(video_url: str) -> str:
    # (This function is complete and correct from the previous version)
    print(f"Tool 'transcribe_youtube_video' (ffmpeg) called with URL: {video_url}")
    video_path, audio_path = None, None
    try:
        os.makedirs(TEMP_DIR, exist_ok=True)
        yt = YouTube(video_url)
        stream = yt.streams.filter(only_audio=True).first()
        video_path = stream.download(output_path=TEMP_DIR)
        audio_path = os.path.join(TEMP_DIR, "output.mp3")
        command = ["ffmpeg", "-i", video_path, "-y", "-q:a", "0", "-map", "a", audio_path]
        subprocess.run(command, check=True, capture_output=True, text=True)
        client = Groq(api_key=os.getenv("GROQ_API_KEY"))
        with open(audio_path, "rb") as audio_file:
            transcription = client.audio.transcriptions.create(file=audio_file, model="whisper-large-v3", response_format="text")
        return str(transcription)
    except Exception as e:
        return f"Error during YouTube transcription: {e}"
    finally:
        if video_path and os.path.exists(video_path): os.remove(video_path)
        if audio_path and os.path.exists(audio_path): os.remove(audio_path)

# --- NEW TOOL Definition: Image Analysis ---
def analyze_image_from_task_id(task_id: str) -> str:
    """
    Downloads an image file for a given task_id and analyzes it using a Vision-Language Model.
    Use this tool ONLY when a question explicitly mentions an image.
    """
    print(f"Tool 'analyze_image_from_task_id' called with task_id: {task_id}")
    try:
        file_url = f"{DEFAULT_API_URL}/files/{task_id}"
        print(f"Downloading image from: {file_url}")
        response = requests.get(file_url)
        response.raise_for_status()
        
        # Initialize the VLM client
        vlm_client = HuggingFaceInferenceAPI(
            model_id="llava-hf/llava-1.5-7b-hf",
            token=os.getenv("HF_TOKEN")
        )
        
        print("Analyzing image with Llava...")
        # The prompt for the VLM needs to be specific.
        # We can just ask it to describe the image in detail.
        text_prompt = "Describe the image in detail."
        result = vlm_client.image_to_text(image=response.content, prompt=text_prompt)
        print(f"Image analysis successful. Result: {result}")
        return result
        
    except Exception as e:
        return f"Error during image analysis: {e}"

# --- Agent Definition ---
class LangChainAgent:
    def __init__(self, groq_api_key: str, tavily_api_key: str, hf_token: str):
        self.llm = ChatGroq(model_name="llama3-70b-8192", groq_api_key=groq_api_key, temperature=0.0)

        self.tools = [
            TavilySearchResults(name="web_search", max_results=3, tavily_api_key=tavily_api_key, description="A search engine for finding up-to-date information on the internet."),
            Tool(name="audio_file_transcriber", func=transcribe_audio_file, description="Use this for questions mentioning an audio file (.mp3, recording). Input MUST be the task_id."),
            Tool(name="youtube_video_transcriber", func=transcribe_youtube_video, description="Use this for questions with a youtube.com URL. Input MUST be the URL."),
            Tool(name="image_analyzer", func=analyze_image_from_task_id, description="Use this for questions mentioning an image. Input MUST be the task_id."),
        ]
        
        prompt = ChatPromptTemplate.from_messages([
            ("system", (
                "You are a powerful problem-solving agent. Your goal is to answer the user's question accurately. "
                "You have access to a web search tool, an audio file transcriber, a YouTube video transcriber, and an image analyzer.\n\n"
                "**REASONING PROCESS:**\n"
                "1.  **Analyze the question:** Determine if a tool is needed. Is it a general knowledge question, or does it mention a specific file type (audio, video, image) or URL?\n"
                "2.  **Select ONE tool based on the question:**\n"
                "    - For general knowledge, facts, or current events: use `web_search`.\n"
                "    - For an audio file, .mp3, or voice memo: use `audio_file_transcriber` with the `task_id`.\n"
                "    - For a youtube.com URL: use `youtube_video_transcriber` with the URL.\n"
                "    - For an image: use `image_analyzer` with the `task_id`.\n"
                "    - For math or simple logic: answer directly.\n"
                "3.  **Execute and Answer:** After using a tool, analyze the result and provide ONLY THE FINAL ANSWER."
            )),
            ("human", "Question: {input}\nTask ID: {task_id}"),
            ("placeholder", "{agent_scratchpad}"),
        ])

        agent = create_tool_calling_agent(self.llm, self.tools, prompt)
        self.agent_executor = AgentExecutor(agent=agent, tools=self.tools, verbose=True, handle_parsing_errors=True)

    def __call__(self, question: str, task_id: str) -> str:
        urls = re.findall(r'https?://[^\s]+', question)
        input_for_agent = {"input": question, "task_id": task_id}
        if urls and "youtube.com" in urls[0]:
            input_for_agent['video_url'] = urls[0]
        try:
            response = self.agent_executor.invoke(input_for_agent)
            return response.get("output", "Agent failed to produce an answer.")
        except Exception as e:
            return f"Agent execution failed with an error: {e}"

# --- Main Application Logic ---
def run_and_submit_all(profile: gr.OAuthProfile | None):
    space_id = os.getenv("SPACE_ID")
    if not profile: return "Please Login to Hugging Face with the button.", None
    username = profile.username
    try:
        groq_api_key = os.getenv("GROQ_API_KEY")
        tavily_api_key = os.getenv("TAVILY_API_KEY")
        hf_token = os.getenv("HF_TOKEN")
        if not all([groq_api_key, tavily_api_key, hf_token]): raise ValueError("An API key (GROQ, TAVILY, or HF) is missing.")
        agent = LangChainAgent(groq_api_key=groq_api_key, tavily_api_key=tavily_api_key, hf_token=hf_token)
    except Exception as e: return f"Error initializing agent: {e}", None
    
    questions_url = f"{DEFAULT_API_URL}/questions"
    try:
        response = requests.get(questions_url, timeout=20)
        response.raise_for_status()
        questions_data = response.json()
    except Exception as e: return f"Error fetching questions: {e}", None
    
    results_log, answers_payload = [], []
    for item in questions_data:
        task_id, q_text = item.get("task_id"), item.get("question")
        if not task_id or not q_text: continue
        answer = agent(question=q_text, task_id=task_id)
        answers_payload.append({"task_id": task_id, "submitted_answer": answer})
        results_log.append({"Task ID": task_id, "Question": q_text, "Submitted Answer": answer})
    
    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
    submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
    submit_url = f"{DEFAULT_API_URL}/submit"
    try:
        response = requests.post(submit_url, json=submission_data, timeout=300)
        response.raise_for_status()
        result_data = response.json()
        final_status = (f"Submission Successful!\nUser: {result_data.get('username')}\n"
                        f"Overall Score: {result_data.get('score', 'N/A')}% "
                        f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
                        f"Message: {result_data.get('message', 'No message received.')}")
        return final_status, pd.DataFrame(results_log)
    except Exception as e: return f"Submission Failed: {e}", pd.DataFrame(results_log)

# --- Gradio Interface ---
with gr.Blocks() as demo:
    gr.Markdown("# Ultimate Agent Runner (Search, Audio, Video, Vision)")
    gr.Markdown("This agent can search, transcribe audio files, transcribe YouTube videos, and analyze images.")
    gr.LoginButton()
    run_button = gr.Button("Run Evaluation & Submit All Answers")
    status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
    results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
    run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])

if __name__ == "__main__":
    print("\n" + "-"*30 + " App Starting " + "-"*30)
    for key in ["GROQ_API_KEY", "TAVILY_API_KEY", "HF_TOKEN"]:
        print(f"✅ {key} secret is set." if os.getenv(key) else f"⚠️ WARNING: {key} secret is not set.")
    print("-"*(60 + len(" App Starting ")) + "\n")
    demo.launch(debug=True, share=False)