import os
import requests
from langchain_core.messages import HumanMessage
from agent import build_graph
from huggingface_hub import hf_hub_download
import pyarrow.parquet as pq
from dotenv import load_dotenv

load_dotenv(override=True)

DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"

graph = build_graph()
resp = requests.get(f"{DEFAULT_API_URL}/questions")
questions = resp.json()

token = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACEHUB_API_TOKEN")
path = hf_hub_download(repo_id='gaia-benchmark/GAIA', filename='2023/validation/metadata.parquet', repo_type='dataset', token=token)
df = pq.read_table(path).to_pandas()
answer_map = dict(zip(df['task_id'], df['Final answer']))

# Check Q1, Q5, Q7
for i in [0, 4, 6]:
    q = questions[i]
    task_id = q['task_id']
    question = q['question']
    ground_truth = answer_map.get(task_id, "NOT FOUND")
    
    result = graph.invoke({"messages": [HumanMessage(content=question)]})
    answer = result['messages'][-1].content
    
    print(f"\n=== Q{i+1} ===")
    print(f"Q: {question[:80]}...")
    print(f"GT: {ground_truth}")
    print(f"Ans: {answer[:50]}")