LangChain OpenSearch RAG

A Retrieval-Augmented Generation (RAG) application using LangChain, OpenSearch, and Ollama for local LLM inference.

Overview

This project implements a document processing and question-answering system that:

1. Processes documents from various formats (PDF, TXT, MD, PY)
2. Generates embeddings using local models via Ollama
3. Stores document chunks and embeddings in OpenSearch
4. Provides a Gradio web interface for document upload and querying
5. Uses Redis for caching to improve performance

Architecture

The system consists of the following components:

• App Service: Python application with Gradio web interface
• OpenSearch: Vector database for storing document embeddings
• OpenSearch Dashboards: Web interface for OpenSearch management
• Redis: Caching layer that saves metadata from processed documents
• Ollama: Local inference for embeddings and LLM responses

Prerequisites

• Docker and Docker Compose
• Ollama installed locally (optional, can use containerized version)
• At least 8GB RAM for running all services
• Git

Quick Start

1. Clone the repository:

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

2. Create a .env file in the infra/secrets directory:

   mkdir -p infra/secrets
   touch infra/secrets/.env

3. Start the services:

   docker-compose -f infra/docker-compose.yml up -d

4. Access the web interface at http://localhost:8081

Configuration

Environment Variables

Key environment variables that can be configured:

Variable	Description	Default
OLLAMA_HOST	URL for Ollama API	http://host.docker.internal:11434
EMBEDDINGS_MODEL	Model for generating embeddings	mxbai-embed-large
LLM_MODEL	Model for text generation	phi3.5
OPENSEARCH_URL	URL for OpenSearch	http://opensearch:9200
REDIS_HOST	Hostname for Redis	redis
CHUNK_SIZE	Size of document chunks	1000
CHUNK_OVERLAP	Overlap between chunks	200

Using Local Ollama

The default configuration uses your local Ollama installation. Make sure Ollama is running:

ollama serve

And that you have the required models:

ollama pull mxbai-embed-large
ollama pull phi3.5

Usage

1. Upload Documents: Use the web interface to upload documents (PDF, TXT, MD, PY)
2. Process Documents: The system will chunk, embed, and store the documents
3. Ask Questions: Query the system about the content of your documents
4. View Sources: See which document chunks were used to generate the answer

Development

Project Structure

langchain-opensearch-rag/
├── infra/                  # Infrastructure configuration
│   ├── docker-compose.yml  # Docker Compose configuration
│   ├── redis/              # Redis configuration
│   └── scripts/            # Helper scripts
├── src/                    # Source code
│   └── app/                # Application code
│       ├── loader/         # Document loading and processing
│       ├── ui/             # Gradio UI components
│       └── main.py         # Application entry point
└── README.md               # This file

Local Development

To develop locally:

1. Create a Python virtual environment:

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

2. Install dependencies:

   pip install -r requirements.txt

3. Run the application:

   python src/app/main.py

Troubleshooting

Common Issues

1. Redis Connection Issues: Ensure Redis is running and accessible from the app container
2. Ollama Connection Issues: Verify Ollama is running and the OLLAMA_HOST is correctly set
3. OpenSearch Connection Issues: Check OpenSearch logs and ensure it's healthy

Logs

To view logs:

# All services
docker-compose -f infra/docker-compose.yml logs -f

# Specific service
docker-compose -f infra/docker-compose.yml logs -f app

License

MIT License