Drift detection, shadow deploys, canaries, rollback and the on-call reality of owning a model.
8h · 8 lessons · 1 challenge
Before this: Inference Optimization & Serving
Models degrade quietly. Nothing errors; the metric just drifts down for six weeks. This module covers the monitoring that catches that — input drift, prediction drift, delayed labels — and the deployment machinery that makes a fix safe to ship.
Includes incident response for ML systems, which is a genuine differentiator in senior interviews.
Six months of production telemetry, replayed. Raise the alarm too early and you cry wolf; too late and the quarter is lost.
Worth 550 XP.
