AI product and engineering lead based in Tokyo, focused on local and on-device AI. I have worked in software engineering since 2001 and I am still hands-on every day: I set the direction, the architecture and the acceptance criteria, work with AI coding agents on the implementation, and own the result.

Houjun Co., Ltd. (Tokyo) — Product Engineer / Representative Director · houjun.dev

• VoxStage — offline multi-voice text-to-speech workstation for Apple Silicon. A local LLM drafts who says each line, the author reviews it, and every generated line is checked against the script before export to audio, subtitles and editing timelines. AGPL-3.0, with a separate commercial licence. · website
• Image MLX Lab — local image generation and editing on Qwen-Image via Apple MLX; mask-based edits leave every pixel outside the selection unchanged. Code MIT; the default model is for non-commercial use. · website
• MedAdapt Lab — open research: can a small language model learn new research findings through continued training, and what does it cost? Five studies with tests frozen before training. · write-up
• Vision Lab — pedestrian detection and multi-object tracking on CPU: YOLOX and ByteTrack in C++ on ONNX Runtime, in Docker, with reproducible MOT17 benchmarks.
• AI-Lab — local-inference studies on one Apple Silicon machine: what ships, what it costs and how it fails.
• llama.cpp — reproducible issue reports on C/C++ audio and media paths: #27693, #27697, #28095.

• Mind Craft Fish — learning app for iPhone, iPad and Mac in four languages, offline-first. · website
• CherryTempo — reminders and shared tasks for iPhone and Mac.

• 1998 — respiratory-sound recognition, my biomedical engineering graduation project, with a university hospital.
• 2004–2012 — co-founded a software company in Nanjing and led distributed real-time video analytics deployed across four metropolitan programs; wrote the core video components in C++.
• 2014 — software lead for a four-wheel-drive robot prototype: real-time obstacle analysis, fast avoidance and remote robotic-arm control.

Chinese (native) · Japanese (JLPT N2) · English (technical reading and writing; working meetings)

Open to applied-AI product and engineering roles.