The scenario
A data platform team needs a relational schema to track ML experiments, training runs, metrics, datasets, model versions and monitoring alerts.
What is in this drawing
Read the decisions behind it.
01
Separate Run and ModelVersion to link training runs to registered model artifacts
02
Use a RunDataset join table to model many-to-many dataset usage across runs
03
Attach alerts to model versions for metric-threshold monitoring
Reusable for ML experiment tracking, model registry backends, and model monitoring dashboards.
More like this