hoduy511/rag-langchain

A production-ready RAG (Retrieval Augmented Generation) system built with FastAPI, LangChain, LangServe, LangSmith, Hugging Face, and Qdrant for document processing and intelligent querying.

RAG LangChain

A production-ready RAG (Retrieval Augmented Generation) system built with FastAPI, LangChain, LangServe, LangSmith, Hugging Face, and Qdrant for document processing and intelligent querying.

Table of Contents

• Core Features
• Tech Stack
• Quick Start
• API Endpoints
• Components
• Development
• Testing
• Configuration

Core Features

• PDF Document Processing with automatic chunking and metadata enrichment
• Vector Search using Qdrant with sentence-transformers/all-MiniLM-L6-v2 embedding model for efficient document retrieval
• Integration with google/flan-t5-base for question answering
• RESTful API with streaming support
• Docker containerization with multi-service architecture

Tech Stack

• FastAPI + Uvicorn
• LangChain + LangServe
• Qdrant Vector Database
• Hugging Face Models (google/flan-t5-base, sentence-transformers/all-MiniLM-L6-v2)
• LangSmith for monitoring and debugging
• PyPDF for document processing

Quick Start

1. Clone the repository:

git clone https://github.com/hoduy511/rag-langchain.git
cd rag-langchain

2. Create and configure your .env file with required variables:

cp .env-dev .env

3. Start the services using make up command:

make up

API Endpoints

Base URLs

• API Documentation: http://localhost:8000/docs
• Interactive Playground: http://localhost:8000/rag/playground

Core Endpoints

• GET /health - Health check endpoint
• POST /api/v1/upload - Upload and process PDF documents
• POST /api/v1/query - Query the knowledge base
• POST /api/v1/search - Perform similarity search

Components

API Layer (src/api/)

• FastAPI application handling HTTP requests
• Route definitions for document upload, querying, and search
• Input validation and response formatting
• CORS and middleware configuration

Core Services (src/services/)

Document Formatter

• Text chunking with configurable size and overlap
• Metadata enrichment
• UTF-8 encoding handling
• Content cleaning and normalization

LLM Service

• google/flan-t5-base model integration
• Text generation pipeline configuration
• Token length management
• Model parameter optimization

PDF Service

• PDF document processing
• Text extraction and cleaning
• Temporary file management
• Chunk generation and storage

Vector Store

• Qdrant vector database integration
• Document embedding using sentence-transformers
• Similarity search functionality
• Collection management and indexing

RAG Chain (src/chains/)

• LangChain implementation for question answering
• Integration with google/flan-t5-base model
• Prompt management and chain composition
• Context retrieval and response generation

Data Models (src/models/)

• Pydantic schemas for request/response validation
• Data transfer object definitions
• Type hints and validation rules

Development

Available make commands for development:

• up: Start all services with docker-compose
• down: Stop all services and remove containers
• logs: View container logs in follow mode
• shell: Open interactive shell in app container
• clean: Remove all containers, volumes and prune system
• test: Run pytest test suite
• format: Format Python code with autopep8 and isort
• lint: Run flake8 linter checks

Testing

The project includes comprehensive tests for all components including API endpoints, RAG chain implementation, and various services.

Configuration

Environment Variables

• API settings
• Qdrant vector database configuration
• Model settings (google/flan-t5-base and sentence-transformers/all-MiniLM-L6-v2)
• LangChain integration parameters
• LangSmith API keys and project settings
• Hugging Face API tokens and model configurations