# Title, CIDR ranges, arrow labels and flow steps for the gallery diagram. Use with: # terravision draw --source gcp-vertex-mlops.tvg.json --annotate gcp-vertex-mlops.annotations.yml format: '0.3' title: Google Cloud Vertex AI MLOps with Workbench, pipelines, a feature store and a private endpoint update: google_compute_subnetwork.workbench: ip_cidr_range: 10.60.0.0/24 google_compute_subnetwork.training: ip_cidr_range: 10.60.16.0/20 google_compute_subnetwork.serving: ip_cidr_range: 10.60.32.0/24 google_vertex_ai_metadata_store.pipelines: label: Vertex AI Pipelines tv_gcp_external_saas.github: label: GitHub google_access_context_manager_service_perimeter.ml: label: VPC-SC Perimeter Ml google_artifact_registry_repository.training_images: label: Artifact Registry Training Images google_bigquery_dataset.prediction_logs: label: BigQuery Prediction Logs google_bigquery_dataset.training: label: BigQuery Training google_cloud_run_v2_service.claims: label: Cloud Run Claims google_cloudbuild_trigger.train: label: Cloud Build Train google_cloudbuildv2_repository.models: label: Repository Models google_compute_forwarding_rule.psc_vertex_ai: label: Forwarding Rule Psc Vertex Ai google_compute_global_forwarding_rule.psc_google_apis: label: Forwarding Rule Psc Google Apis google_iap_client.workbench: label: IAP Workbench google_kms_key_ring.ml: label: Cloud KMS Ml google_logging_project_sink.ml: label: Log Sink Ml google_secret_manager_secret.github_token: label: Secret Manager Github Token google_storage_bucket.artifacts: label: Cloud Storage Artifacts google_vertex_ai_endpoint.fraud: label: Vertex AI Endpoint Fraud google_vertex_ai_feature_online_store.features: label: Feature Store Features google_vertex_ai_model.fraud: label: Model Registry Fraud google_workbench_instance.notebooks: label: Workbench Notebooks connect: google_iap_client.workbench: - google_workbench_instance.notebooks: IAP google_vertex_ai_endpoint.fraud: - google_bigquery_dataset.prediction_logs: Request-response logging flows: Train: description: A merged change retrains the model color: '#D93025' steps: - resource: tv_gcp_users_icon.data_scientists -> google_iap_client.workbench xlabel: Sign in detail: Data scientists reach their Workbench notebooks through Identity-Aware Proxy - resource: tv_gcp_server.github -> google_cloudbuildv2_repository.models xlabel: Merge detail: Merging their change to main notifies Cloud Build - resource: google_cloudbuild_trigger.train -> google_artifact_registry_repository.training_images xlabel: Build detail: which builds the training image into Artifact Registry - resource: google_cloudbuild_trigger.train -> google_vertex_ai_metadata_store.pipelines xlabel: Run detail: and starts a Vertex AI pipeline on workers in the training subnet - resource: google_vertex_ai_metadata_store.pipelines -> google_compute_global_forwarding_rule.psc_google_apis xlabel: Data detail: The pipeline reads features from BigQuery and writes artifacts to Cloud Storage inside the perimeter - resource: google_vertex_ai_metadata_store.pipelines -> google_vertex_ai_model.fraud xlabel: Register detail: and registers the evaluated model in the Model Registry Serve: description: A claim is scored color: '#188038' steps: - resource: google_vertex_ai_model.fraud -> google_vertex_ai_endpoint.fraud xlabel: Deploy detail: An approved model version is deployed to the fraud endpoint - resource: tv_gcp_users_icon.claims_handlers -> google_cloud_run_v2_service.claims xlabel: Claim detail: Claims handlers open a claim in the claims app - resource: google_cloud_run_v2_service.claims -> google_compute_forwarding_rule.psc_vertex_ai xlabel: Score detail: which calls the endpoint through its Private Service Connect address - resource: google_vertex_ai_endpoint.fraud -> google_vertex_ai_feature_online_store.features xlabel: Features detail: The endpoint looks up the customer's features in the online store - resource: google_vertex_ai_endpoint.fraud -> google_bigquery_dataset.prediction_logs xlabel: Log detail: and logs requests and responses to BigQuery for model monitoring