The scenario
A recommender system logs every inference request and recommendation, captures user feedback, snapshots features, and computes monitoring metrics with drift alerts.
What is in this drawing
Read the decisions behind it.
01
Separate InferenceRequest, Recommendation, and FeedbackEvent to isolate serving, ranking, and ground truth
02
Track ModelVersion and FeatureSnapshot to correlate drift with deployment and input distribution
03
Use MonitoringMetric and DriftAlert to store computed KPIs and operational actions
Use this ERD when designing a data store for real-time ML inference monitoring, especially for recommender or ranking systems requiring feedback loops and drift detection.
More like this