app.py

from fastapi import FastAPI, UploadFile, File
import pandas as pd
import faiss
import pickle
import os
import json
from sentence_transformers import SentenceTransformer
from exa_py import Exa
from groq import Groq

app = FastAPI()

# =============================
# 🔑 KEYS
# =============================
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
EXA_API_KEY = os.getenv("EXA_API_KEY")

client = Groq(api_key=GROQ_API_KEY)
exa = Exa(api_key=EXA_API_KEY)

# =============================
# 🧠 LOAD MODELS
# =============================
embed_model = SentenceTransformer('all-mpnet-base-v2')

index = faiss.read_index("faiss_index.index")

with open("startup_texts.pkl", "rb") as f:
    startup_texts = pickle.load(f)

# =============================
# 🔍 RETRIEVAL
# =============================
def retrieve_similar(problem, k=3):
    vec = embed_model.encode([problem], convert_to_numpy=True)
    distances, indices = index.search(vec, k)

    return [
        {"text": startup_texts[idx], "score": float(distances[0][i])}
        for i, idx in enumerate(indices[0])
    ]

# =============================
# 🌐 WEB SEARCH
# =============================
def search_web(query):
    try:
        response = exa.search(query, num_results=5)
        return [r.text or r.summary or "" for r in response.results]
    except:
        return []

# =============================
# 🤖 QWEN (Groq)
# =============================
def ask_qwen(prompt):
    completion = client.chat.completions.create(
        model="qwen/qwen3-32b",
        messages=[
            {"role": "system", "content": "You are a strict fact-checking analyst. Return ONLY valid JSON."},
            {"role": "user", "content": prompt}
        ],
        temperature=0.3,
        max_tokens=512
    )
    return completion.choices[0].message.content

# =============================
# 🧠 PIPELINE (FIXED)
# =============================
def analyze_problem(problem):

    retrieved = retrieve_similar(problem)

    if retrieved and retrieved[0]["score"] < 2.0:
        context = "\n\n".join([r["text"] for r in retrieved])
    else:
        web = search_web(problem)
        context = "\n\n".join(web[:3])

    prompt = f"""
Problem:
{problem}

Evidence:
{context}

Return ONLY valid JSON in this format:

{{
  "status": "SOLVED or UNSOLVED",
  "reason": "one short sentence",
  "gaps": ["gap1", "gap2"],
  "new_problem": "rewrite of the problem"
}}
"""

    response = ask_qwen(prompt)

    try:
        parsed = json.loads(response)
    except:
        parsed = {
            "status": "ERROR",
            "reason": "invalid JSON from model",
            "gaps": [],
            "new_problem": ""
        }

    parsed["problem"] = problem

    return parsed

# =============================
# 🚀 API
# =============================
@app.post("/analyze")
async def analyze(file: UploadFile = File(...)):

    df = pd.read_csv(file.file)

    results = []

    for p in df.iloc[:, 0].tolist():
        try:
            results.append(analyze_problem(p))
        except Exception as e:
            results.append({
                "problem": p,
                "status": "ERROR",
                "reason": str(e),
                "gaps": [],
                "new_problem": ""
            })

    return {
        "success": True,
        "results": results
    }