Построение MLOps-пайплайна от данных до деплоя

data

Ml pipeline is a Claude Code skill that designs and implements production-grade ML pipeline infrastructure: configures experiment tracking with MLflow or Weights & Biases, builds Kubeflow Pipelines and Airflow DAGs for training orchestration, sets up feature store schemas with Feast, deploys model registries, and automates retraining and validation workflows. Reach for it when you need a reproducible end-to-end training pipeline, when experiments keep diverging because nothing is tracked, when distributed training jobs need proper scheduling, or when you want automatic rollback to the best model version without manual intervention. The skill generates Python code with ready-made templates for MLflow, Kubeflow, Airflow, and Prefect, integrates data validation via Great Expectations, versions datasets with DVC, pins dependencies and random seeds, and keeps credentials out of source code. The direction is MLOps and training orchestration, not model inference or production serving — for that, see the cloud-architect or kubernetes-specialist skills. Works well for teams building ML infrastructure on Kubernetes or cloud environments; not the right fit when you just need a quick prototype with no infrastructure requirements — python-pro covers that case.