main.py

import os
import zipfile
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel

from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_groq import ChatGroq
from langchain.chains import RetrievalQA
from langchain.prompts import PromptTemplate

app = FastAPI()

# === Globals ===
llm = None
embeddings = None
vectorstore = None
retriever = None
chain = None

class QueryRequest(BaseModel):
    question: str

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

    # 1) Init LLM & Embeddings
    llm = ChatGroq(
        model="meta-llama/llama-4-scout-17b-16e-instruct",
        temperature=0,
        max_tokens=1024,
        api_key=os.getenv("API_KEY"),
    )
    embeddings = HuggingFaceEmbeddings(
        model_name="intfloat/multilingual-e5-large",
        model_kwargs={"device": "cpu"},
        encode_kwargs={"normalize_embeddings": True},
    )

    # 2) Unzip & Load both FAISS vectorstores
    # — First index
    zip1 = "faiss_index.zip"
    dir1 = "faiss_index"
    if not os.path.exists(dir1):
        with zipfile.ZipFile(zip1, 'r') as z:
            z.extractall(dir1)
        print("✅ Unzipped FAISS index 1.")
    vs1 = FAISS.load_local(
        dir1,
        embeddings,
        allow_dangerous_deserialization=True
    )
    print("✅ FAISS index 1 loaded.")

    # — Second index
    zip2 = "faiss_index(1).zip"
    dir2 = "faiss_index_extra"
    if not os.path.exists(dir2):
        with zipfile.ZipFile(zip2, 'r') as z:
            z.extractall(dir2)
        print("✅ Unzipped FAISS index 2.")
    vs2 = FAISS.load_local(
        dir2,
        embeddings,
        allow_dangerous_deserialization=True
    )
    print("✅ FAISS index 2 loaded.")

    # 3) Merge them
    vs1.merge_from(vs2)
    vectorstore = vs1
    print("✅ Merged FAISS indexes into a single vectorstore.")

    # 4) Create retriever & QA chain
    retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
    prompt = PromptTemplate(
        template="""
You are an expert assistant on Islamic knowledge.
Use **only** the information in the “Retrieved context” to answer the user’s question.
Do **not** add any outside information, personal opinions, or conjecture—if the answer is not contained in the context, reply with “لا أعلم”.
Be concise, accurate, and directly address the user’s question.

Retrieved context:
{context}

User’s question:
{question}

Your response:
""",
        input_variables=["context", "question"],
    )
    chain = RetrievalQA.from_chain_type(
        llm=llm,
        chain_type="stuff",
        retriever=retriever,
        return_source_documents=False,
        chain_type_kwargs={"prompt": prompt},
    )
    print("✅ QA chain ready.")

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

@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))