Software Engineering

A world that never logged off Twilight was an LPMUD - a text-based multiplayer world I spent hours through my teens and my early 20s. How I agonized over character deaths, getting banned by the trigger-happy arch back then. The MUD has gone through many hands, but David (with the help of ZH) has kept it alive on AWS. The mudlib underneath it dates to 1996. The math has felt wrong, it felt wrong when David first made me a wiz during my uni days but it was just difficult to go through the codebase. Recently David started GPT-6 Astra on the old codebase, and it kept flagging how badly connected the areas were. When the copy of the code was sent over, Fable 5.1 started pointing out how badly written the skills code was. No surprises there, honestly, most of the old generation of wizzes were not particuarly great coders, nor did they have the tooling to succeed. Line by line editing using ed, anyone? ...

Introduction In my 2025 in Review, I noted that 2025 was the year I switched to AI-assisted coding, and that organizations which embrace it will push far ahead of those which don’t. That was written in late December, with the ink still wet on the experience. Four months later, with a few more projects under the belt, it’s time to revisit that claim. Three projects form the evidence base: Bridge Ratings - (May 2025 - presented at Data Science SG in Jun 2025) Admissions Prediction - straightforward binary classification, but with shifting requirements Vaultshift - started January 2026, a bridge results platform that has grown from a single-session MVP into a full multi-session system integrated with automated masterpointing and ratings from above It’s interesting to see how things have changed really fast over the past year. ...