app.py

import streamlit as st
from PyPDF2 import PdfReader
from langchain.vectorstores.cassandra import Cassandra
from langchain.indexes.vectorstore import VectorStoreIndexWrapper
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.llms import HuggingFaceHub
from langchain.text_splitter import CharacterTextSplitter
import cassio
from dotenv import load_dotenv
import os

load_dotenv()

ASTRA_DB_APPLICATION_TOKEN = os.getenv("ASTRA_DB_APPLICATION_TOKEN")
ASTRA_DB_ID = os.getenv("ASTRA_DB_ID")
HUGGINGFACEHUB_API_TOKEN = os.getenv("HUGGINGFACEHUB_API_TOKEN")

# === Streamlit UI Setup ===
st.set_page_config(page_title="Query PDF with Free Hugging Face Models", layout="wide")
st.title("📄💬 Query PDF using LangChain + AstraDB (Free Hugging Face Models)")

# === File Upload ===
uploaded_file = st.file_uploader("Upload your PDF", type=["pdf"])

if uploaded_file:
    st.success("✅ PDF uploaded successfully!")
    process_button = st.button("🔄 Process PDF")

    if process_button:
        # Initialize AstraDB
        cassio.init(token=ASTRA_DB_APPLICATION_TOKEN, database_id=ASTRA_DB_ID)

        # Read PDF contents
        pdf_reader = PdfReader(uploaded_file)
        raw_text = ""
        for page in pdf_reader.pages:
            content = page.extract_text()
            if content:
                raw_text += content

        # Split text into chunks
        text_splitter = CharacterTextSplitter(
            separator="\n", chunk_size=800, chunk_overlap=200, length_function=len
        )
        texts = text_splitter.split_text(raw_text)

        # === Embeddings ===
        embedding = HuggingFaceEmbeddings(
            model_name="sentence-transformers/all-MiniLM-L6-v2"
        )

        # === Hugging Face LLM ===
        llm = HuggingFaceHub(
            repo_id="mistralai/Mistral-7B-Instruct-v0.1",
            huggingfacehub_api_token=HUGGINGFACEHUB_API_TOKEN,
            model_kwargs={"temperature": 0.5, "max_new_tokens": 512}
        )

        # === Create vector store and index ===
        vector_store = Cassandra(
            embedding=embedding,
            table_name="qa_mini_demo",
            session=None,
            keyspace=None,
        )
        vector_store.add_texts(texts[:50])
        st.success(f"📚 {len(texts[:50])} chunks embedded and stored in AstraDB.")

        astra_vector_index = VectorStoreIndexWrapper(vectorstore=vector_store)

        # === Ask Questions ===
        st.header("🤖 Ask a question about your PDF")
        user_question = st.text_input("💬 Type your question here")

        if user_question:
            with st.spinner("Thinking..."):
                answer = astra_vector_index.query(user_question, llm=llm).strip()
                st.markdown(f"### 🧠 Answer:\n{answer}")

                st.markdown("### 🔍 Top Relevant Chunks")
                docs = vector_store.similarity_search_with_score(user_question, k=4)
                for i, (doc, score) in enumerate(docs, 1):
                    st.markdown(f"**Chunk {i}** — Relevance Score: `{score:.4f}`")
                    st.code(doc.page_content[:500], language="markdown")