app.py

import gradio as gr
import pandas as pd
import httpx
import re, json, asyncio
from openai import OpenAI
from itertools import cycle
import aiohttp
import asyncio
import requests
import csv
import os

# 🔑 Hardcoded credentials (you provided)
OPENAI_API_KEY = "OAPI1"
DEEPSEEK_API_KEY = "DAPI"
GOOGLE_API_KEY = "API1"
GOOGLE_CSE_ID = "CX1"

OPENAI_MODEL = "gpt-4o-mini"
DEEPSEEK_MODEL = "deepseek-chat"

# ---------------- Helper functions ----------------

async def google_cse(query, num=4):
    url = "https://www.googleapis.com/customsearch/v1"
    params = {"q": query, "cx": GOOGLE_CSE_ID, "key": GOOGLE_API_KEY, "num": num}
    async with httpx.AsyncClient(timeout=30.0) as client:
        r = await client.get(url, params=params)
        r.raise_for_status()
        items = r.json().get("items", [])
        return [{"title": it.get("title"), "link": it.get("link"), "snippet": it.get("snippet")} for it in items]

async def call_openai(messages):
    url = "https://api.openai.com/v1/chat/completions"
    headers = {"Authorization": f"Bearer {OPENAI_API_KEY}"}
    payload = {"model": OPENAI_MODEL, "messages": messages, "temperature": 0, "max_tokens": 800}
    async with httpx.AsyncClient(timeout=60.0) as client:
        r = await client.post(url, headers=headers, json=payload)
        r.raise_for_status()
        return r.json()["choices"][0]["message"]["content"]

async def call_deepseek(messages):
    url = "https://api.deepseek.com/v1/chat/completions"
    headers = {"Authorization": f"Bearer {DEEPSEEK_API_KEY}"}
    payload = {"model": DEEPSEEK_MODEL, "messages": messages, "temperature": 0, "max_tokens": 800}
    async with httpx.AsyncClient(timeout=60.0) as client:
        r = await client.post(url, headers=headers, json=payload)
        r.raise_for_status()
        return r.json()["choices"][0]["message"]["content"]

def split_claims(text):
    parts = re.split(r"[.;\n]", str(text))
    return [p.strip() for p in parts if len(p.strip()) > 5][:10]

async def verify_claim(claim):
    results = await google_cse(claim, num=3)
    sources = "\n".join([f"- {r['title']} ({r['link']}): {r['snippet']}" for r in results])
    user = f"CLAIM: {claim}\n\nSOURCES:\n{sources}"
    messages = [
        {"role": "system", "content": "You are a fact-checker. Return JSON {status, reason, citations[]}"},
        {"role": "user", "content": user}
    ]
    oai, dsk = await asyncio.gather(call_openai(messages), call_deepseek(messages))
    return {"claim": claim, "openai": oai, "deepseek": dsk, "sources": results}

async def process_csv(file):
    df = pd.read_csv(file)
    if "Requirement" not in df.columns and "Requirements" not in df.columns:
        raise gr.Error("CSV must have a 'Requirement' or 'Requirements' column")
    col = "Requirement" if "Requirement" in df.columns else "Requirements"

    all_results = []
    for idx, row in df.iterrows():
        text = str(row[col])
        claims = split_claims(text)
        claim_results = [await verify_claim(c) for c in claims]
        all_results.append({"ID": row.get("ID", idx), "results": claim_results})

    # Make summary table
    summary = [{"ID": r["ID"], "Claims": len(r["results"])} for r in all_results]
    table = pd.DataFrame(summary)

    # JSON output
    json_bytes = json.dumps(all_results, indent=2, ensure_ascii=False).encode("utf-8")
    return table, json_bytes

# ---------------- Gradio UI ----------------

with gr.Blocks() as demo:
    gr.Markdown("# 🔎 Citation Checker (OpenAI + DeepSeek + Google CSE)\nUpload your CSV and check requirements automatically.")
    file_in = gr.File(label="Upload CSV", file_types=[".csv"])
    run_btn = gr.Button("Run Verification", variant="primary")
    table_out = gr.Dataframe(label="Summary")
    json_out = gr.File(label="Download JSON Results")

    run_btn.click(fn=process_csv, inputs=file_in, outputs=[table_out, json_out])

if __name__ == "__main__":
    demo.launch()