Tak Shing Chan

I spent 15 years doing theoretical physics research — the kind of work where you spend months on a problem before you know if your approach even converges, and where being rigorous about assumptions matters more than being fast. That instinct — question the model before you trust the result — turned out to transfer directly to machine learning.

I'm now moving into applied ML engineering: fine-tuning transformer models, building and deploying real inference services (not just notebooks), and shipping projects end-to-end — from a Kaggle dataset to a live, containerized API. MLworld is where I document that work in public.

Outside of ML projects, I also run arc6ai, an AI automation consultancy — which is where a lot of the production/deployment habits (Docker, Cloud Run, real APIs instead of scripts) come from.