app.py

import os
import streamlit as st
import numpy as np
import faiss
from groq import Groq
from pydrive.auth import GoogleAuth
from pydrive.drive import GoogleDrive
from sentence_transformers import SentenceTransformer

# Constants
DRIVE_FILE_LINK = "https://drive.google.com/file/d/1kYGomSibXW-wCFptEMcWP12jOz1390OK/view?usp=drive_link"
GROQ_MODEL = "llama-3.3-70b-versatile"

# Authentication and setup for Google Drive
@st.cache_resource
def load_drive_content(file_link):
    gauth = GoogleAuth()
    gauth.LocalWebserverAuth()
    drive = GoogleDrive(gauth)
    file_id = file_link.split('/d/')[1].split('/view')[0]
    downloaded_file = drive.CreateFile({'id': file_id})
    downloaded_file.GetContentFile("document.pdf")
    return "document.pdf"

# Chunking and embedding creation
@st.cache_resource
def prepare_embeddings(document_path):
    from PyPDF2 import PdfReader
    
    reader = PdfReader(document_path)
    text = ""
    for page in reader.pages:
        text += page.extract_text()

    # Create chunks of 500 characters with a sliding window of 200
    chunk_size = 500
    chunk_overlap = 200
    chunks = [text[i:i+chunk_size] for i in range(0, len(text), chunk_size - chunk_overlap)]
    
    # Embedding model
    embedder = SentenceTransformer("all-MiniLM-L6-v2")
    embeddings = embedder.encode(chunks, convert_to_tensor=True).detach().numpy()

    # Store in FAISS
    vector_dim = embeddings.shape[1]
    index = faiss.IndexFlatL2(vector_dim)
    index.add(embeddings)
    return chunks, index

# Groq setup
@st.cache_resource
def groq_client():
    return Groq(api_key=os.environ.get("GROQ_API_KEY"))

# Retrieve and query vector DB
def query_vector_db(query, chunks, index, embedder):
    query_embedding = embedder.encode([query], convert_to_tensor=True).detach().numpy()
    D, I = index.search(query_embedding, k=1)  # Find top result
    if I[0][0] != -1:  # Valid match
        return chunks[I[0][0]]
    return "No relevant content found."

# Streamlit application
def main():
    st.title("RAG-based Application with Groq")

    # Load document and prepare FAISS
    st.info("Loading document and preparing FAISS...")
    document_path = load_drive_content(DRIVE_FILE_LINK)
    chunks, index = prepare_embeddings(document_path)
    embedder = SentenceTransformer("all-MiniLM-L6-v2")
    client = groq_client()
    
    # Interface
    user_input = st.text_input("Enter your query:")
    if user_input:
        context = query_vector_db(user_input, chunks, index, embedder)
        st.write("**Relevant Context:**", context)

        # Query Groq model
        with st.spinner("Querying Groq model..."):
            chat_completion = client.chat.completions.create(
                messages=[
                    {"role": "user", "content": f"Based on this context: {context}, {user_input}"}
                ],
                model=GROQ_MODEL,
            )
            st.write("**Groq Model Response:**", chat_completion.choices[0].message.content)

if __name__ == "__main__":
    main()