Model management, feature extraction, and evaluation services
Manage TF SavedModel lifecycle — register versions, assign tags (prod/dev/test), browse artifacts. Timestamp-based versioning with exclusive tag system.
Define and extract ML features from raw data. Feature registry with lineage tracking, group management, model-feature mapping, and per-request access analytics.
Run TF model inference end-to-end — loads models (LRU cached), extracts features, runs prediction, logs results. Full observability with latency and history.
Versioned business rules per model — create, activate, deactivate, and evaluate rules. Apply post-prediction logic like confidence thresholds and overrides.
Submit and monitor Vertex AI training pipeline runs. View run history, check status, and link to GCP Console for detailed pipeline monitoring.
ClearML experiment dashboard — track training metrics, compare runs, visualize hyperparameters, and monitor pipeline performance across experiments.