RAG Demo - Spring AI

A demonstration of RAG (Retrieval-Augmented Generation) technology using Spring AI with Ollama and PGVector.

📋 Overview

This project demonstrates the implementation of a RAG (Retrieval-Augmented Generation) system that allows users to ask questions about PDF documents and receive contextualized answers using artificial intelligence. The system uses a vector database to store and search for relevant information within the documents.

Current Configuration

• Analyzed Document: "La Fortune des Rougon" by Émile Zola
• Chat Model: Llama 3.1 8B via Ollama
• Embedding Model: nomic-embed-text
• Vector Database: PGVector (PostgreSQL)

🛠️ Technologies Used

• Java 21
• Spring Boot 3.5.3
• Spring AI 1.0.0
• Ollama (for LLM models)
• PostgreSQL with PGVector extension
• Docker Compose (for infrastructure)

📦 Main Dependencies

• spring-ai-starter-model-ollama - Integration with Ollama
• spring-ai-pdf-document-reader - PDF reading and processing
• spring-ai-starter-vector-store-pgvector - Vector database
• spring-ai-starter-model-chat-memory - Conversation memory management
• spring-ai-advisors-vector-store - Advisors for vector search

🚀 Prerequisites

1. Java 21 or higher
2. Maven 3.6+
3. Docker and Docker Compose
4. Ollama installed locally

Installing Ollama

# macOS
brew install ollama

# Start the service
ollama serve

# Download required models
ollama pull llama3.1:8b
ollama pull nomic-embed-text

🏃‍♂️ Quick Start

1. Clone the repository

git clone <repository-url>
cd spring-ai-RAG-demo

2. Start PostgreSQL with PGVector

cd src/main/docker
docker-compose up -d

3. Check Ollama service

curl http://localhost:11434/api/tags

4. Build and Run the Application

mvn clean compile

# First run: fill the vector store and next continue with console interaction
mvn spring-boot:run -Dspring-boot.run.arguments="--fillVectorStore"

# Next runs: start the application with console interaction
mvn spring-boot:run

💬 Usage

Once the application is started, you can interact with it via the console:

Ask a question: Who are the main characters of the novel? ?

[The AI responds based on the content of the document...]

Ask a question: exit

Example questions

• "What is the historical context of the novel?"
• "Describe the main character"
• "What are the main themes of the work?"
• "Summarize the first chapter"

⚙️ Configuration

Changing the source document

To analyze another PDF document, edit the application.yml file:

rag:
  system-prompt: "You are an expert on the provided document and you answer questions based on the information given"
  document-path: "classpath:your-document.pdf"

Changing the chat model

spring:
  ai:
    ollama:
      chat:
        options:
          model: llama3.2:3b  # Lower memory model
          # model: llama3.1:8b  # Default model
          temperature: 0.1

Database Configuration

Ensure the database connection settings in application.yml match your PostgreSQL setup:

spring:
  datasource:
    url: jdbc:postgresql://localhost:5432/vector_db
    username: postgres_user
    password: postgres_password

🏗️ Architecture

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Document PDF  │────│  Text Splitter  │────│  Embeddings     │
└─────────────────┘    └─────────────────┘    └─────────────────┘
                                                        │
┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│  User Question  │────│   Chat Client   │────│   PGVector DB   │
└─────────────────┘    └─────────────────┘    └─────────────────┘
                                │
                       ┌─────────────────┐
                       │   Ollama LLM    │
                       └─────────────────┘

🔧 Development

Project Structure

src/
├── main/
│   ├── java/com/zenika/demo/rag/
│   │   └── RagDemoApplication.java      # Application entry point
│   ├── resources/
│   │   ├── application.yml              # Configuration
│   │   └── *.pdf                        # Documents source
│   └── docker/
│       └── compose.yml                  # Docker Compose for PostgreSQL with PGVector

Main Features

1. Document Ingestion: Reading and splitting PDFs into chunks
2. Embeddings Generation: Converting text into vectors
3. Vector Storage: Saving into PGVector
4. Semantic Search: Retrieving relevant context
5. Answer Generation: Using an LLM to respond

🐛 Troubleshooting

Common Issues

1. Ollama service not running

   # Check if Ollama is running
   ollama serve

2. Database connection issues

   # Restart PostgreSQL with PGVector
   docker-compose down && docker-compose up -d

3. Models not found

   # Download required models
   ollama pull llama3.1:8b
   ollama pull nomic-embed-text

4. Unsufficient memory for model

   • Use a smaller model (llama3.2:3b)
   • Increase memory: -Xmx4g

Checking Logs

mvn spring-boot:run

# Check PostgreSQL logs
docker-compose logs pgvector

# List Ollama models
ollama list

📚 References

• Spring AI Documentation
• Ollama
• PGVector
• Spring Boot

🤝 Contribution

This project is an educational demonstration. Contributions are welcome to:

• Add new document types
• Improve system prompts
• Optimize performance
• Add tests