thecodemonkey / SimilaritySearch

This project is part of a tutorial demonstrating how to use Kotlin or Java with Spring-AI to generate embeddings and perform simple similarity searches.

Similarity Search with Spring AI

Demo | tutorial

Table of Contents

• Similarity Search with Spring AI
  • Table of Contents
  • Introduction
  • Project structure
  • Requirements
  • Quick Start
    • Clone Project
    • Build & Run
    • run applicaiton
    • run in docker
    • use the UI
  • Troubleshooting
    • Memory Usage
    • Windows WSL2

Introduction

This project is part of a tutorial demonstrating how to use Kotlin or Java with Spring-AI to generate embeddings and perform simple similarity searches. It provides a minimal implementation approach. This sample project contains both a Kotlin and a Java implementation. I like to use Kotlin because the syntax is much leaner and more precise than the classic Java. But it's up to you which one you prefer.

Project structure

root
 |
 |-- src
       |-- java                         # java source code
       |-- kotlin                       # kotlin source code                   
       |-- resources           
                  |-- data              # raw and vecorized data
                  |-- model             # optional local model
                  |-- static            # frontend implementation of PoC
                  |-- application.yml   

Requirements

• jdk 21
• gradle (current version)
• min 2GB free ram

Quick Start

Clone Project

git clone https://github.com/thecodemonkey/SimilaritySearch.git
cd SimilaritySearch

Build & Run

Using Gradle:

./gradlew build

run applicaiton

Once the application is running, the API is available at http://localhost:7070. I have explicitly used a different port to avoid collisions with applications that are already running. The configuration can be found in application.yml/server.port

Example request:

./gradlew bootRun
curl http://localhost:7070/embeddings?text=give%20me%20love%20songs

run in docker

The application can also be started in Docker. The already generated image from the GitLab Container Registry can be used, or the local project can be rebuilt as a Docker image and run locally.

run from ghcr.io:

docker run -p 7070:7070 ghcr.io/thecodemonkey/similaritysearch:main

# arm

docker run --platform linux/amd64 -p 80:7070 ghcr.io/thecodemonkey/similaritysearch:main

use the UI

open the URL http://localhost:7070/ in your browser to see and use the ui. use the input field for the search. You can use the microphone icon for the voice function. Note which language is selected. You can currently select German or English.

Troubleshooting

Memory Usage

The application requires slightly more memory than usual. If you get an error message with “out of memory” or something with “java heap size” then you should increase the JAVA memory.

java JAVA_OPTS="-Xmx1024m" -jar similarity-search.jar

Windows WSL2

If you are working with WSL under windows, note that wsl has its own IP, which changes with every restart. This is how you can determine the IP from the WSL console

inside wsl terminal:

ip addr show eth0

grab the specific ip and open in browser on your host machine:

curl http://172.20.45.2:7070