Built a research-grade predictive modeling workflow using XGBoost, point-in-time data controls, validation gates, and calibrated decision logic. The work focuses on reproducible forecasting and auditable model quality rather than unsupported performance claims.