DriftAudit

DriftAudit is a web app (with optional Slack alerts) that continuously monitors your production data streams and surfaces “silent” distribution shifts using unsupervised learning. Instead of requiring labeled outcomes, it profiles feature distributions, embeddings, and correlation structure over time, then flags statistically meaningful drift and segments where it’s happening (e.g., a new customer cohort, a new device type, a specific geography). It generates plain-English incident reports, shows which features moved, and provides example records to speed up debugging. It also supports baseline snapshots (by season, campaign, or release) so you don’t get spammed by expected changes. This is an AI app combined with a traditional monitoring product: the value is in reliable drift detection, clear explanations, and workflow integration—not flashy dashboards. Realistically, it won’t replace full MLOps suites, but it can be a focused, cheaper alternative for teams that just need drift visibility fast.