app.py

import streamlit as st

# ✅ Correct imports (new structure)
from langchain_community.document_loaders import PyPDFLoader, TextLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_community.llms import HuggingFacePipeline

from langchain.chains import RetrievalQA
from transformers import pipeline


# -------------------------------
# Load Documents
# -------------------------------
def load_documents(uploaded_files):
    documents = []
    for file in uploaded_files:
        with open(file.name, "wb") as f:
            f.write(file.getbuffer())

        if file.name.endswith(".pdf"):
            loader = PyPDFLoader(file.name)
        else:
            loader = TextLoader(file.name)

        documents.extend(loader.load())
    return documents


# -------------------------------
# Split Documents
# -------------------------------
def split_documents(documents):
    splitter = RecursiveCharacterTextSplitter(
        chunk_size=500,
        chunk_overlap=50
    )
    return splitter.split_documents(documents)


# -------------------------------
# Create Vector Store
# -------------------------------
def create_vectorstore(chunks):
    embeddings = HuggingFaceEmbeddings(
        model_name="sentence-transformers/all-MiniLM-L6-v2"
    )
    return FAISS.from_documents(chunks, embeddings)


# -------------------------------
# Load Local LLM (FREE)
# -------------------------------
def load_llm():
    pipe = pipeline(
        "text2text-generation",   # ✅ FIXED
        model="google/flan-t5-base",
        max_length=512
    )
    return HuggingFacePipeline(pipeline=pipe)


# -------------------------------
# Build QA Chain (with strict prompt)
# -------------------------------
def build_qa(vectorstore):
    llm = load_llm()
    retriever = vectorstore.as_retriever()

    qa = RetrievalQA.from_chain_type(
        llm=llm,
        retriever=retriever
    )
    return qa


# -------------------------------
# Streamlit UI
# -------------------------------
st.set_page_config(page_title="RAG Chatbot", layout="wide")
st.title("📄 Chat with Your Documents (RAG)")

uploaded_files = st.file_uploader(
    "Upload PDF or TXT files",
    accept_multiple_files=True
)

if uploaded_files:
    with st.spinner("Processing documents..."):
        docs = load_documents(uploaded_files)
        chunks = split_documents(docs)
        vectorstore = create_vectorstore(chunks)
        qa_chain = build_qa(vectorstore)

    st.success("Documents ready!")

    query = st.text_input("Ask a question from your documents")

    if query:
        with st.spinner("Generating answer..."):
            result = qa_chain.run(query)
            st.write("### Answer:")
            st.write(result)