Retrieval-Augmented Generation (RAG) with MongoDB and Spring AI: Bringing AI to Your Java Applications

Welcome to the RAG App GitHub repository! This project demonstrates how to build a Retrieval-Augmented Generation (RAG) system using Spring Boot, MongoDB Atlas, and OpenAI. With RAG, you can use your own data to supplement the responses generated by a large language model (LLM), ensuring more accurate, relevant, and up-to-date answers.

What is Retrieval-Augmented Generation?

Retrieval-Augmented Generation (RAG) is a technique that combines vector search and large language models (LLMs) to generate context-aware answers based on proprietary or external data that was not part of the model’s initial training. RAG consists of three main components:

1. Pre-trained LLM: We use OpenAI's GPT model to generate responses.
2. Vector search: We retrieve relevant documents from a MongoDB database.
3. Vector embeddings: Numerical representations of your data that capture semantic meaning.

Key Features

• Integrates with MongoDB Atlas for vector search capabilities.
• Uses OpenAI for embeddings and generating smart, context-driven answers.
• Implements a simple vector store and document embedding service.
• Includes REST endpoints for document loading and AI-powered question-answering.

Prerequisites

To run this project, you need:

1. Java 21 or higher.
2. Maven (for dependency management).
3. MongoDB Atlas (M10+ cluster required for vector search).
4. OpenAI API key.

Make sure you have these tools installed and accounts configured before proceeding.

Setup Instructions

1. Clone the repository

git clone https://github.com/timotheekelly/RagApp.git
cd RagApp

2. Configure the project

Edit the application.properties file with your MongoDB URI and OpenAI API key.

spring.application.name=RagApp  
spring.ai.openai.api-key=<Your-API-Key>
spring.ai.openai.chat.options.model=gpt-4o  
spring.ai.vectorstore.mongodb.initialize-schema=true  
spring.data.mongodb.uri=<Your-Connection-URI>
spring.data.mongodb.database=rag

3. Build the project

Use Maven to build the project.

mvn clean install

4. Run the application

Start the Spring Boot application:

mvn spring-boot:run

5. Load documents

Navigate to the following endpoint to load documents into the vector store:

http://localhost:8080/api/docs/load

6. Ask a question

You can ask questions by sending a request to this endpoint:

http://localhost:8080/question?message=Your question here

For example:

http://localhost:8080/question?message=How to analyze time-series data with Python and MongoDB? Explain the steps

Key Components

• EmbeddingModel: Configured to use OpenAI for generating document embeddings.
• VectorStore: Utilizes MongoDB Atlas for storing and retrieving vectors for similarity searches.
• DocsLoaderService: Reads and embeds documents, then stores them in the MongoDB vector store.
• RagController: Handles question-answering requests by performing vector searches and interacting with the LLM.

Conclusion

This project demonstrates how to integrate a retrieval-augmented generation system with MongoDB Atlas and OpenAI to enhance Java applications. By combining vector search and generative AI, you can create smart, context-aware applications tailored to your own data.

Resources

• MongoDB Developer Center
• Spring Boot Documentation
• OpenAI API