agent.py

import os
import requests
from openai import OpenAI

class GaiaAgent:
    def __init__(self):
        self.client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
        self.instructions = (
            "You are a highly skilled and concise research assistant solving GAIA benchmark questions.\n"
            "Analyze attached files, extract relevant information, reason step-by-step internally,\n"
            "and return only the final factual answer in the correct format. Avoid explanations."
        )
        self.api_url = "https://agents-course-unit4-scoring.hf.space"

    def fetch_file_content(self, task_id: str) -> str:
        try:
            url = f"{self.api_url}/files/{task_id}"
            response = requests.get(url, timeout=10)
            response.raise_for_status()
            content_type = response.headers.get("Content-Type", "")
            if "text" in content_type or "csv" in content_type or "json" in content_type:
                return response.text[:3000]  # Truncate to 3000 chars
            elif "application/pdf" in content_type:
                return "[PDF detected. Summarize manually if needed.]"
            elif "image" in content_type:
                return "[Image detected. Describe the image if needed.]"
            elif "audio" in content_type:
                return "[Audio detected. Provide transcription if needed.]"
            else:
                return f"[Unsupported file type: {content_type}]"
        except Exception as e:
            return f"[Error fetching file: {e}]"

    def __call__(self, question: str, task_id: str = None) -> str:
        file_context = ""
        if task_id:
            file_context = self.fetch_file_content(task_id)
            if file_context:
                file_context = f"Attached File Context:\n{file_context}\n"

        # Add scratchpad-like structure
        prompt = (
            f"{self.instructions}\n\n"
            f"{file_context}"
            f"Question: {question}\n"
            f"Think step-by-step to extract relevant facts and solve the task.\n"
            f"Final Answer (no explanation, just the answer):"
        )

        response = self.client.chat.completions.create(
            model="gpt-4-turbo",
            messages=[
                {"role": "system", "content": self.instructions},
                {"role": "user", "content": prompt}
            ],
            temperature=0.0,
        )
        return response.choices[0].message.content.strip()