Building Junas forced me to hit the ground running with architecting scalable and evaluable ML pipelines that survive CI-enforced benchmarking, but more importantly taught me how to scope (and cut) product features based on real user impact and needs.

Implementing more features just because engineering is fun and the tech itself is cool can be tempting, but when that leads to building 'a solution looking for a problem', its likely you're building something noone would actually use.

Junas now lives at https://github.com/gongahkia/junas for those interested.