What problem does it solve? Machine learning models degrade as real-world data shifts, and manually retraining, validating, and promoting models is error-prone and slow. This Skill designs automated retraining pipelines that safely retrain models, compare challengers against champions, and roll back instantly when production metrics regress. ## Core Features & Use Cases - Retraining Orchestration: Build Prefect flows that fetch recent data, validate quality, engineer features, and train challengers with the champion's hyperparameters. - Champion vs. Challenger Evaluation: Compare models on a held-out test set with minimum improvement thresholds and automated promote/reject decisions via MLflow. - Promotion & Rollback: Automate MLflow stage transitions with approval gates, shadow deployment guidance, and instant rollback to archived versions. - Use Case: A data science team notices prediction drift in their order-propensity model. Use this Skill to design a drift-triggered retraining pipeline that retrains on the last 60 days of data, validates quality, and auto-promotes only if AUC improves. ## Quick Start Design an automated retraining pipeline for my order-propensity model that retrains weekly, compares the challenger against the production champion on a holdout set, and requires manual approval before promotion.
