Production ML Deployment, Drift Monitoring & Auto-Retraining
MLOps Pipeline & Drift Guardrail Telemetry Cockpit
Select a deployed production model, inject live feature drift (PSI / KL-Divergence) using the simulator slider, and inspect how our Python/FastAPI inference router automatically shifts traffic to the fallback ensemble while triggering an asynchronous retraining DAG.
p99 Inference Latency
14.2 ms
SLO < 35ms (Healthy)
Feature Drift (PSI)
0.08
Threshold: 0.20
1. Active Production Endpoints
2. Simulate Covariate Drift (PSI)
Nominal (0.08)
Drag PSI above 0.20 to test automated canary circuit-breaking and Prefect/Airflow DAG retraining triggers.
0.02 (Baseline)
0.20 (Alert Gate)
0.45 (Severe Drift)
3. Automated MLOps Pipeline DAG Status
run_id: mlops_2026_10_27_0419STEP 01 • FEAST
Feature Validation
GreatExpectations schema pass (1.2M rows)
STEP 02 • EVIDENTLY
Drift Monitor
PSI 0.08 < 0.20 gate
STEP 03 • MLFLOW
Shadow Retrain
Standby (on-demand)
STEP 04 • ARGO ROLLOUTS
Traffic Router
v4.2 serving 92% / v4.1 8%