Notes

How our team went from a time-consuming problem to a working MVP in a week, with a pitch grounded in facts and a demo that made the value clear.

How I put a Flutter team's AI coding agents behind a pre-execution hook and a CI gate, what real incidents shaped the rules, and what a red-team model and a human reviewer found when they tried to break it.

A Dart framework that turns a ticket's requirements and definition of done into acceptance criteria, then holds the PR until a validator has driven the app and attached screenshots and recordings.

DojoEasy suggests who's in a class photo. The model runs in the browser, the face templates stay on the instructor's device, and two brothers in the same gi shaped the matching logic.

Our crash rate on older Android phones was 10% before the SKIMS x NIKE drop. This is how the app handles product images now, from the CDN URL down to the disk cache.

Step timings pointed at 304 Flutter engines booting per run, caches the tag builds could never read, and a staging job that compiled the same Dart twice.