ShiftSentry

ShiftSentry is a web app (with optional Slack alerts) that monitors data drift and anomaly patterns in production ML pipelines without requiring a full observability platform. You connect it to a warehouse table or feature store export, define “golden” reference windows, and it continuously checks for distribution shifts, missingness spikes, schema changes, and weird value ranges that quietly degrade model performance. It produces plain-English incident summaries with likely root causes (upstream job failure, new category explosion, bot traffic, ETL type coercion) and suggests concrete checks to add. The product is intentionally narrow: it’s not training models, not APM, not a full MLOps suite. It’s a lightweight drift/anomaly watchdog that small teams can actually deploy and maintain. Expect some false positives early; the value is reducing time-to-diagnosis when something changes at 2 a.m. and your metrics lag behind reality.