I build production AI and software systems, and the part I care most about is the unglamorous one: what happens when they change.
I am an engineer in the semiconductor industry by background (BEng Mechanical), and I design, ship and operate software alongside that, including LLM features running in production.
-
Production model migration case study: a verified production LLM model migration, kept strictly separate from the stronger method that came after it, and how that method evolved into ModelPromote. Runnable offline.
-
ModelPromote: governs the switch when you change the AI model in production. Approval, read-back activation, telemetry verification, rollback, one portable record. TypeScript, zero runtime dependencies, on npm with verified provenance.
-
asinstalled: tests an npm package the way its users receive it: packed, installed into a clean project, and run on every Node version its
enginesfield promises. Built for my own releases; it runs in ModelPromote's CI.
- Peer-reviewed review of natural-fibre bicycle brake calipers (2017), from my engineering degree at Multimedia University.