app.py

# app.py - FINAL, 100% WORKING
import os
import subprocess
import gradio as gr
from langchain_community.vectorstores import Weaviate
from langchain.chains import RetrievalQA
from langchain.prompts import PromptTemplate
from sentence_transformers import SentenceTransformer
import weaviate
import time
import requests

# === FIX: Correct Groq import ===
from langchain_groq import ChatGroq

# === SECRETS ===
WEAVIATE_URL = os.environ["WEAVIATE_URL"]
WEAVIATE_KEY = os.environ["WEAVIATE_KEY"]
os.environ["GROQ_API_KEY"] = os.environ["GROQ_API_KEY"]

# === WAIT FOR WEAVIATE TO BE READY ===
def wait_for_weaviate(url, key, timeout=120):
    headers = {"Authorization": f"Bearer {key}"}
    ready_url = f"{url}/v1/.well-known/ready"
    print("Waiting for Weaviate to be ready...")
    for i in range(timeout):
        try:
            response = requests.get(ready_url, headers=headers)
            if response.status_code == 200:
                print("Weaviate is ready!")
                return True
        except:
            pass
        print(f"Attempt {i+1}/{timeout}... waiting 1s")
        time.sleep(1)
    print("Weaviate did not start in time.")
    return False

# === AUTO-INGEST ON START ===
def run_ingestion():
    print("Starting ingestion...")
    if not wait_for_weaviate(WEAVIATE_URL, WEAVIATE_KEY):
        print("Cannot connect to Weaviate. Skipping ingestion.")
        return

    result = subprocess.run(["python", "ingest.py"], capture_output=True, text=True)
    print(result.stdout)
    if result.returncode != 0:
        print("Ingestion failed:", result.stderr)
    else:
        print("Ingestion complete!")

# Run once
run_ingestion()

# === RAG CHAIN (NO @gr.cache) ===
def get_rag_chain():
    client = weaviate.Client(
        url=WEAVIATE_URL,
        auth_client_secret=weaviate.AuthApiKey(WEAVIATE_KEY)
    )
    embedder = SentenceTransformer("all-MiniLM-L6-v2")
    vectorstore = Weaviate(client, "Paper", "text", embedding=embedder)
    retriever = vectorstore.as_retriever(search_kwargs={"k": 3})

    llm = ChatGroq(model="llama-3.1-70b-instruct", temperature=0)
    prompt = PromptTemplate.from_template(
        "Answer using only this context:\n{context}\n\nQuestion: {question}\nAnswer:"
    )
    return RetrievalQA.from_chain_type(
        llm=llm, retriever=retriever, chain_type_kwargs={"prompt": prompt}
    )

def search(query):
    try:
        qa = get_rag_chain()
        result = qa.invoke({"query": query})
        answer = result["result"]
        sources = "\n\n".join([
            f"**{doc.metadata.get('title', 'No Title')}**\n{doc.page_content[:300]}..."
            for doc in result["source_documents"]
        ])
        return answer, sources or "No sources found."
    except Exception as e:
        return f"Error: {str(e)}", "Check logs."

# === GRADIO UI ===
with gr.Blocks(title="ArXiv RAG") as demo:
    gr.Markdown("# ArXiv RAG Search")
    gr.Markdown("10K+ papers • Llama-3.1 • Weaviate")

    txt = gr.Textbox(label="Ask", placeholder="What is attention?", lines=2)
    btn = gr.Button("Search", variant="primary")
    out1 = gr.Markdown(); out2 = gr.Markdown()

    btn.click(search, txt, [out1, out2])

demo.launch()