What is this project?

An enterprise-grade MLOps pipeline that ingests raw CRM data, applies a feature engineering layer (normalisation, ordinal encoding, interaction terms), trains an XGBoost classifier, and pushes churn probability scores to a PostgreSQL dashboard consumed by the CRM team. Great Expectations validates data quality at each stage; Prefect orchestrates the schedule; MLflow tracks every experiment with full model registry support.

The standout feature is drift detection: Kolmogorov–Smirnov tests run nightly on each feature. When any feature drifts beyond a configurable threshold, Prefect automatically triggers a retraining job with fresh data. In production, the model stayed accurate 3 months beyond the initial training cutoff thanks to this feedback loop — a critical win for a SaaS client with 50K active users.