ShiftProof

ShiftProof is a web app (with optional CLI) that monitors supervised learning models in production by tracking feature distribution shifts and label/target stability, then recommends concrete actions. You connect it to your data warehouse or event stream, define your model’s expected schema, and it automatically computes drift metrics, segment-level changes, and “what changed” explanations. When drift crosses thresholds, it generates a prioritized checklist: retrain now vs. wait, which segments are impacted, and which features likely caused the shift. It also keeps a lightweight “model card” history so teams can correlate drift events with releases, upstream ETL changes, or seasonality. This is not a magic auto-ML platform; it’s a pragmatic early-warning system that reduces firefighting and silent accuracy decay. The MVP focuses on tabular classification/regression first, because that’s where most business models live and where drift hurts quietly.