main.py

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from langchain.chains import RetrievalQA
from langchain.prompts import PromptTemplate
from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_groq import ChatGroq
import zipfile
import os

app = FastAPI()

# === Startup config ===
class QueryRequest(BaseModel):
    question: str

llm = None
retriever = None
chain = None

import os
os.environ["TOKENIZERS_PARALLELISM"] = "false"

@app.on_event("startup")
def load_components():
    global llm, retriever, chain

    api_key = os.getenv('api_key')

    # --- Load LLM ---
    llm = ChatGroq(
        model="meta-llama/llama-4-scout-17b-16e-instruct",
        temperature=0,
        max_tokens=1024,
        api_key=api_key
    )

    # --- Load Embeddings ---
    embeddings = HuggingFaceEmbeddings(
        model_name="intfloat/multilingual-e5-large",
        model_kwargs={"device": "cpu"},
        encode_kwargs={"normalize_embeddings": True},
    )

    # --- Unzip Vectorstore if needed ---
    zip_path = "faiss_index.zip"
    extract_path = "faiss_index"
    if not os.path.exists(extract_path):
        with zipfile.ZipFile(zip_path, 'r') as z:
            z.extractall(extract_path)
        print("✅ Unzipped FAISS index.")

    # --- Load FAISS Vectorstore & create retriever ---
    vectorstore = FAISS.load_local(
        extract_path,
        embeddings,
        allow_dangerous_deserialization=True
    )
    retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
    print("✅ FAISS index loaded.")

    # --- Prepare prompt template ---
    quiz_solving_prompt = """
You are an Arabic Hadith Finder assistant.
Your goal is to provide an accurate and concise answer extracted directly from the provided retrieved context.
Your task is to output only the exact Arabic Hadith (as it appears in the context), removing any extraneous or irrelevant data.

Instructions:
1. Identify the segment in the retrieved context that directly answers the user's question.
2. Output the Hadith exactly as it appears in Arabic in the context.
3. Remove any information that does not pertain directly to the query.
4. If the context does not contain sufficient information to answer the question, respond with "لا أعلم". Do not add or infer any extra information.
5. Provide the complete reference of the Hadith (if available), including:
   - Chapter Number and Name 
   - Section Number and Name
   - Hadith Number
   - Arabic Isnad and Matn
   - Arabic Grade (if present)
   - Hadith Book name

Retrieved context:
{context}

User's question:
{question}

Your response:
"""
    prompt = PromptTemplate(
        template=quiz_solving_prompt,
        input_variables=["context", "question"]
    )

    # --- Assemble a stateless RetrievalQA chain (no memory) ---
    chain = RetrievalQA.from_chain_type(
        llm=llm,
        chain_type="stuff",
        retriever=retriever,
        return_source_documents=False,
        chain_type_kwargs={"prompt": prompt},
        verbose=False,
    )

@app.get("/")
def root():
    return {"message": "Arabic Hadith Finder API is up..."}

@app.post("/query")
def query(request: QueryRequest):
    try:
        result = chain.invoke({"query": request.question})
        return {"answer": result["result"]}
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))