Intermediate MLOps & Production
30 MIN READ
ML Pipeline Orchestration: Temporal, Airflow, Kubeflow, Ray — Which Layer Does What
A precise mental model for ML pipeline orchestration—mapping durable backend workflows (Temporal), data schedulers (Airflow, Prefect, Dagster), ML-native pipeline frameworks (Kubeflow, Metaflow, ZenML), and distributed compute engines (Ray). Built for engineers who need to answer 'design an ML pipeline' in interviews. Includes 2025-2026 updates: Airflow 3, KFP v2, Ray 2.x, MLflow 3.