MLOps for AI
For ML and software engineers who can train a model but have never had to run one in production. You will version code, data and models, track experiments, promote and roll back through a registry, and ship to managed cloud platforms or Kubernetes with the monitoring that tells you when a model has gone wrong.
Course Content
Make a model reproducible from code, data, config and seed, ship it as a pinned container through GitOps, and automate retraining behind gates that stop a worse model. 3 lessons, about 45 minutes.
Record every run so it can be compared and audited, version data with DVC, run pipelines on Kubeflow, and promote and roll back models through MLflow aliases. 3 lessons, about 45 minutes.
Weigh SageMaker, Vertex AI and the Hugging Face Hub on cost and lock-in, then run models on Kubernetes with probes, autoscaling, KServe and prediction monitoring. 2 lessons, about 30 minutes.
Build a churn-prediction system end to end: tracked training, a threshold chosen with money, a registered champion, a container on Kubernetes, alerts and a rehearsed rollback. 1 lesson, about 20 minutes.