domino-model-monitoring

To detect data drift for production models deployed on Domino, you need to enable monitoring on your Domino Model API and verify that drift metrics appear in the monitoring dashboard. It requires a deployed model API and training data for baselining.

Model health monitoring tracks production model quality and detects data drift to prevent degraded performance. By triggering alerts when feature shifts occur, it enables rapid remediation workflows and triggers retraining pipelines to maintain accuracy.

You do not need ground-truth labels strictly required to track model quality, as the system supports prediction capture and quality tracking with optional ground-truth integration. However, adding ground truth enhances accuracy monitoring for your Domino endpoints.

You can automate remediation and retraining pipelines when model drift is detected by configuring alerts through the monitoring configuration. This requires access to monitoring settings to trigger automated workflows for production ML models.

Prerequisites to start monitoring a Domino Model API include a deployed model API endpoint, training data for baselining drift detection, and access to the monitoring configuration to enable alerts and automated remediation workflows.