I'm Jinu-coffee roaster by profession.

I roast coffee by day and build oddly specific software by night. This site is where I log my dev experiments—grip detection, overlay managers, ergonomic workflows, and whatever else I end up chasing.

I’m here to learn out loud, reflect when things break, and share the process in case it helps someone else.

I’ve recently finished Roastmaster, a bespoke (love that word) machine-learning program for a 540lb probat/burns roaster. Everyday as a roaster, environmental factors change - room temp & humidity, bean type, temp & age, etc. These factors all matter to a roaster, and a lot of times, it’s down to experience and ehem very good guessing. Well I know everyone is sick of the word AI and ML, but at times like these when a minimum batch is 200lbs (that’s a lot of money down the drain if the roast doesn’t match our profile), we want to take the best data we can and build a robust model that we can compare our experience with.