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 top-tier research assistant for the GAIA benchmark. "
            "You analyze documents, reason step by step, and always provide a single, concise, and correct answer. "
            "If a file is provided, extract all relevant information. Use only information from the question and file. "
            "Show your reasoning before the answer, but end with 'Final Answer: <your answer>'."
        )
        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=15)
            response.raise_for_status()

            content_type = response.headers.get("Content-Type", "")
            if any(t in content_type for t in ["text", "csv", "json"]):
                return response.text[:6000]  # Allow more context for better answers
            elif "application/pdf" in content_type:
                return "[PDF file detected. Use a PDF parser to extract text.]"
            else:
                return f"[Unsupported file type: {content_type}]"
        except Exception as e:
            return f"[Error downloading or reading 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"Here is the related file content:\n{file_context}\n"

        prompt = (
            f"{self.instructions}\n\n"
            f"{file_context}"
            f"Question: {question}\n"
            "Show your reasoning step by step, then provide the final answer as 'Final Answer: <answer>'."
        )

        response = self.client.chat.completions.create(
            model="gpt-4o",  # Use the latest, most capable model for better accuracy
            messages=[
                {"role": "system", "content": self.instructions},
                {"role": "user", "content": prompt}
            ],
            temperature=0.0,
            max_tokens=1024,
        )

        return response.choices[0].message.content.strip()