batch-scoring

MLOps project that recommends movies to watch implementing Data Engineering and MLOps best practices.

Reproducible credit-default modeling in Python with PostgreSQL ingestion, XGBoost, SHAP explanations, calibration analysis, simulated lending policies and portable batch scoring.

Flags healthcare claims likely to deny before submission, cutting simulated rework cost 38.5% on held-out future claims. Leakage-safe DuckDB SQL feature layer, calibrated sklearn models compared on a time split, cost-based decision threshold, batch CLI scoring 45k claims/sec. Synthetic data only, no PHI.

Scheduled GitHub Actions pipeline that scores customers with a churn model and writes a high-risk report

Production-style batch ML pipeline for delivery lead-time risk scoring using Python, SQL and BigQuery.