3D ATC game · demo
Hong Kong Dream International Airport
A browser-based 3D air traffic control game set in Hong Kong.
I turn messy requirements into specs, agent workflows, and working software — then I verify before I ship.
3D ATC game · demo
A browser-based 3D air traffic control game set in Hong Kong.
3D walkthrough · live
Pok Fu Lam 3D walkthrough with time-of-day and fly mode. Captured with Locapack.
2.5D remake · three.js · live
A warm toy-diorama remake of the 1986 NES classic Circus Charlie, with remastered piano music and mobile-friendly play. Supports a physical controller. Performance optimized for both phone and desktop. Built with Grok Bot.
three.js · interactive science
An interactive education app: peel a 3D Jupiter to explore the cloud deck, the Great Red Spot, and the dilute core found by NASA's Juno mission. Best viewed on mobile devices for the 3D effect.
3D walkthrough · live
Interactive Victoria Harbour 3D walkthrough (WASD, lights, time of day). The door opens with a Google Authenticator code.
Built with Grok Build · live
Explorable 3D map of Ocean Park Hong Kong — Waterfront, Summit, cable car and Ocean Express.
Built with Grok Build · live
Portrait remake of the 1983 Bandai LSI Double Play (伊賀対甲賀). Local Wi‑Fi 2P: same room, two phones, same network.
I run coding agents in parallel (Cursor, Claude Code, Grok Bot / Grok Build) to turn plain-language requirements into specs, working apps, tests and review. I build reusable skills, prompts, templates and connectors so non-technical people can use the same stack, with guardrails, evals and human-in-the-loop checks before anything ships. Utilising AI isn’t only raising one person’s productivity — managing agents turns you into a manager of capacity (~10-person output from one hire). I don’t run one-shot prompts. I run agents in feedback loops: plan → build → test → critique → retry, with handoffs when the work splits. Failures feed the next pass (evals, diffs, human checkpoints) so it converges before anything ships.
I ship with agent feedback loops and Python/React. I keep Boeing IE habits — map the work, write what done means, stay through UAT — and I use prompt eval when the role needs it, not as the headline.
Score LLM outputs against written guidelines; revise prompts; retest. GenAI / prompt evaluation.
Python, FastAPI, Django, React, PostgreSQL, Docker, REST/JSON, Git
787 Travelers; 747 Business Planning; 787 Change & Refurb. Requirements with crews/vendors, process mapping, UAT/adoption.

Highly-ranked coding bootcamp in the US.

University of Washington
Web apps and games built with AI.