Kohei Ozaki — smly

Software Engineer

Works

Game AI

RiichiEnv: Research Environment for Mahjong AI

A high-performance research environment for Riichi Mahjong AI development. The core game logic is implemented in Rust for fast state transitions and rollouts, with a Python Gym-style API designed for reinforcement learning. Features include compatibility with the Mortal bot via the MJAI protocol, flexible rule configuration (Tenhou/MJSoul presets, 3-player/4-player modes), and an integrated replay viewer for Jupyter Notebooks.

smly/RiichiEnv

RiichiLab: Mahjong AI Competition Platform

An open arena for Riichi Mahjong AI where developers can pit their bots against each other 24/7 via WebSocket connections. Participants authenticate with GitHub OAuth, register a bot, and connect using the provided API to play full four-player games in real time. The game server is built on RiichiEnv's Rust simulator. Bots receive serialized Observation objects over the WebSocket and return Action responses, sharing the same interface as the local RiichiEnv library for seamless development. The platform features an OpenSkill-based rating system, detailed game logs with key-event shortcuts for analysis, and a replay viewer with permalink support for debugging.

riichi.dev

AI Competitions

I have been actively competing in AI/ML competitions since 2011, spanning over 15 years. Ranked as high as 4th globally on Kaggle (out of 490K+ users) with 23 gold medals.

I actively participate in competitions co-located with top academic conferences and am particularly drawn to those involving satellite imagery analysis.

Beyond competitions, I also engage in community activities and have been serving as a Google Developer Expert for Kaggle since 2025.

Satellite Imagery

I have a strong interest in remote sensing technologies and their applications, including satellite imagery analysis, UAV-based small-object detection, and LiDAR measurements, as well as geolocalization techniques.

Academic Conferences

Industry

Kohei Ozaki · MMXXVI