Projects

Watches order books for anomalies as they happen, flags suspicious wallets at the entity level, and scores risk with a trained model — with a separate pipeline that audits the detector's own calls, rather than trusting it blindly.

A self-hosted app that ranks live auction listings by what a specific person actually wants — starts from rules, then learns from what they click and skip.

The Docker Compose stack running my household's media, calendar, and automation — separate networks per service, least-privilege containers, and a README that explains the "why" behind each call, not just the "what."