Automated Retraining Trigger AI Prompt

Design an automated model retraining system that triggers based on monitored signals.

1. Retraining trigger conditions (any one is sufficient):
   - Performance degradation: model accuracy on recent data drops below {{performance_threshold}}
   - Data drift: PSI > 0.2 for any top-10 feature by importance
   - Prediction drift: KS test p-value < 0.05 on prediction distribution vs baseline
   - Scheduled: time-based trigger every {{retrain_schedule}} (e.g. weekly, monthly)
   - New data volume: {{new_data_threshold}} new labeled samples available since last training

2. Trigger detection pipeline:
   - Run drift checks daily as a scheduled job
   - Log trigger signals to a monitoring database
   - When a trigger fires: log which signal, the metric value, and the threshold exceeded

3. Retraining execution:
   - Submit training job to compute cluster (Kubernetes Job, Airflow DAG, or SageMaker Pipeline)
   - Use the latest full dataset (not just new data) with a sliding window if dataset grows unbounded
   - Run with the same config as the current production model to enable fair comparison

4. Model promotion gate:
   - New model must beat current production model on a fixed evaluation set by > {{min_improvement}}%
   - If gate passes: automatically promote to staging, trigger deployment pipeline
   - If gate fails: alert the ML team, do not auto-promote

5. Human-in-the-loop option:
   - For high-stakes models: require human approval before any promotion, even if gate passes

Return: drift detection script, trigger condition implementation, retraining job submission code, and promotion gate logic.