DriftGuard

DriftGuard is a web app (with optional Slack alerts) that monitors production datasets and ML model inputs for data drift, schema changes, and silent pipeline failures. It connects to common warehouses and feature stores, profiles key columns, and tracks statistical shifts over time (PSI, KS test, missingness, cardinality, outliers). When something changes, it generates a plain-English incident report, likely root causes (new upstream source, ETL bug, seasonality), and a prioritized checklist of fixes. It also keeps an audit trail so teams can prove monitoring existed for compliance. This is a combination traditional + AI app: traditional monitoring/metrics plus an LLM layer that summarizes incidents, suggests remediation, and drafts Jira tickets. Be realistic: this won’t beat enterprise platforms on breadth, but it can win by being lightweight, fast to adopt, and priced for small teams that still suffer expensive model/pipeline outages.