MLOps Pipeline for Predictive Maintenance
06/2026 — PresentProduction-grade MLOps workflow for a machine-failure classifier (AI4I 2020, ~10,000 sensor records): training, experiment tracking and a model registry in MLflow, with automatic champion promotion.
- Raised PR-AUC with XGBoost from 0.38 baseline to 0.83; CI enforces PR-AUC ≥ 0.75 before every promotion.
- Deployed via FastAPI, data-drift monitoring with Evidently, containerised with Docker Compose — 38 tests, ~90% coverage.