DocChat: Langchain Retrieval System

This Streamlit application implements a Langchain-based retrieval system for processing PDF documents and conducting conversational retrieval using Langchain's capabilities.

• Read More: here

RAG Architecture

Streamlit UI

Overview

DocChat is a Langchain-based retrieval system that processes PDF documents and creates a conversational retrieval chain. It leverages multiple technologies to extract text, generate embeddings, and enable chat-based querying over processed content.

Tech Stack

• FastAPI – Serves as the backend API for processing PDFs and handling chat requests.
• Streamlit – Provides the frontend user interface for uploading PDFs and interacting with the conversational system.
• Langchain – A core library for NLP tasks such as text splitting and conversational retrieval.
• Google Palm & Google Generative Language – Used for generating embeddings.
• FAISS – Facebook AI Similarity Search used for efficient similarity search over embeddings.
• PyMuPDF (fitz) – Extracts text from PDFs.
• Docker & Docker Compose – Containerizes and orchestrates the backend and frontend applications.
• Python-dotenv – Loads environment variables (e.g. API keys) from a .env file.

Project Setup

Prerequisites

1. Python Environment: Python 3.x is required.
2. Environment Variables: Create a .env file in the project root with the content:

   GOOGLE_API_KEY=your_google_api_key_here
   

   Replace your_google_api_key_here with your actual Google API key.

Installation (Local Setup)

1. Clone the Repository:

   git clone https://github.com/Varunv003/langchain-palm2-rag_application

2. Set Up Virtual Environment:

   python -m venv venv
   # On Windows:
   .\venv\Scripts\activate
   # On macOS/Linux:
   source venv/bin/activate

3. Install Dependencies:

   pip install -r requirements.txt

4. Initialize Folder Structure (if needed):

   python template.py

5. Running the Streamlit App (Frontend):

   streamlit run app.py

   The application will be available at http://localhost:8501.
6. Running the FastAPI App (Backend):

   uvicorn main:app --reload --host 0.0.0.0 --port 8000

   The backend API will be available at http://localhost:8000.

Running with Docker

This project includes Dockerfiles for both the FastAPI backend and the Streamlit frontend. Docker Compose is used to orchestrate both services.

Building and Running via Docker Compose

1. Ensure Docker Desktop is Running.
2. From the project root (where the docker-compose.yml is located), run:

   docker-compose up --build

3. Services:
   • Backend (FastAPI) will be available at http://localhost:8000.
   • Frontend (Streamlit) will be available at http://localhost:8501.

Docker Compose File Structure

Your docker-compose.yml defines two services:

• backend: Built using Dockerfile.backend and exposing port 8000.
• frontend: Built using Dockerfile.frontend and exposing port 8501, with a dependency on the backend service.

Usage

• Upload PDFs: Use the sidebar in the Streamlit interface to upload PDF files.
• Process Documents: Click "Submit and Process" to extract text, generate embeddings, and initialize the conversational chain.
• Chat: Ask questions related to the processed PDFs through the chat interface. The backend retrieves and forms responses using the Langchain conversational chain.

Future Improvements

• Enhance error handling and user feedback.
• Optimize scalability and performance for larger documents.
• Integrate additional AI models or refine existing conversational models for improved responses.