app.py

from fastapi import FastAPI, Request, Form
from fastapi.responses import HTMLResponse, JSONResponse
from fastapi.staticfiles import StaticFiles
from fastapi.templating import Jinja2Templates

import os
import shutil
from tqdm import tqdm

def load_pdf_and_split(pdf_path):
    loader = PyPDFLoader(pdf_path)
    documents = loader.load()
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=2500, chunk_overlap=400)
    return text_splitter.split_documents(documents)

# LangChain
from langchain_community.document_loaders import PyPDFLoader
from langchain_community.vectorstores import FAISS
from langchain_community.chat_models import ChatOpenAI
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import RetrievalQA
from langchain_community.embeddings import HuggingFaceEmbeddings

# ====================
# FastAPI setup
# ====================
app = FastAPI()
app.mount("/static", StaticFiles(directory="."), name="static")
templates = Jinja2Templates(directory=".")

# ====================
# Limpiar FAISS anterior
# ====================
if os.path.exists("faiss_index"):
    shutil.rmtree("faiss_index")

# ====================
# Cargar documentos PDF
# ====================
split_docs = load_pdf_and_split("1 مساعد ممارس ملف المحور.pdf")

# Debug
for i, doc in enumerate(split_docs[:5]):
    print(f"Chunk {i+1}:\n{doc.page_content[:300]}\n{'-'*40}")

# ====================
# Embeddings multilingües
# ====================
model_name = "asafaya/bert-base-arabic"
embeddings = HuggingFaceEmbeddings(model_name=model_name)

# ====================
# Crear FAISS vectorstore
# ====================
vectorstore = FAISS.from_documents(split_docs, embeddings)
vectorstore.save_local("faiss_index")  # Opcional, para caching

# ====================
# QA Chain
# ====================
qa_chain = RetrievalQA.from_chain_type(
    llm=ChatOpenAI(temperature=0.2, model_name="gpt-4o-mini"),
    chain_type="refine",
    retriever=vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 5})
)

# ====================
# Rutas FastAPI
# ====================
@app.get("/", response_class=HTMLResponse)
async def read_root(request: Request):
    return templates.TemplateResponse("index.html", {"request": request})

@app.post("/preguntar")
async def preguntar(request: Request, pregunta: str = Form(...)):
    respuesta = qa_chain.run(pregunta)
    return JSONResponse({"respuesta": respuesta})

# ====================
# Ejecutar localmente
# ====================
if __name__ == "__main__":
    import uvicorn
    uvicorn.run("app:app", host="0.0.0.0", port=7860)