Private project

Boardgames AI Lab

Game-AI platform for MCTS, self-play, training and model evaluation

Boardgames AI Lab is a local game-AI and backend lab for Tic-Tac-Toe, Connect Four, Onitama and Chess. Java/Spring Boot models rules, PUCT/MCTS, self-play and arena evaluation, while a React/Vite interface makes games, jobs, training and model artifacts operable.

Boardgames AI Lab Play Studio with a running Connect Four match

Overview

The project shows backend and AI engineering beyond classic CRUD systems: rules, state models, search algorithms, training data and a model registry have to work together reproducibly. The current version uses Java/Spring Boot for orchestration and rules, Python/PyTorch for BoardResNet training and a React/Vite operator UI, replacing the earlier Vaadin interface.

Engineering highlights

Rule models and game states for Tic-Tac-Toe, Connect Four, Onitama and Chess

PUCT/MCTS, self-play, arena evaluation and job orchestration as traceable backend components

React/Vite operator UI for games, simulations, training and model registry

BoardResNet training with Python/PyTorch, ONNX export and reproducible model artifacts

Screenshots and demos

Boardgames AI Lab dashboard with game library and runtime metrics — Dashboard for local game-AI engineering: playable games, active jobs, registered models and runtime state.

Boardgames AI Lab training lab with BoardResNet configuration — Training Lab for BoardResNet configuration, preflight, local trainer selection and reproducible initialization.