18-year-old AI-native builder and independent researcher in Hong Kong. Previously at Kimi (Moonshot AI).
I explore what AI agents actually do to software over time—where they help, where they fail, and how we can evaluate the codebases they leave behind. I also build small products to test those ideas in the real world.
Website · Writing · X · YouTube · Email
- DriftBench — a benchmark for measuring code entropy resistance during long-term, agent-driven software evolution.
- ScholarClaw — an open research automation engine that turns a plain-text direction into experiments, a paper draft, and Overleaf-ready output.
- DSE Speaking — an AI oral-exam assistant designed around the real Hong Kong DSE speaking workflow.
- File2Web — a small tool that turns text into a clean, shareable web page.
Before focusing on agents and AI-native products, I shipped iOS apps including EmoEase, Niuma Clock, and Emotion Galaxy. Together they reached the App Store charts, more than 50,000 downloads, and a Swift Student Challenge Distinguished Winner award. They are proof of how I learned to ship—not my current focus.