A model that works in a notebook isn't a product yet. MLOps is the set of practices that turns experiments into reliable systems: tracking experiments and versioning models, building repeatable pipelines, serving predictions through an API or a batch job, and keeping everything tested, automated and monitored once it is live.
You'll start with the full MLOps lifecycle, from experiment tracking and orchestration to deployment, monitoring and CI/CD. Then you'll look at serving large language models at scale, including containers, Kubernetes and high-throughput inference engines such as vLLM.
These courses are long, so treat them as a reference: watch the parts you need, then come back for the rest. When you're done, continue on to Safety & Observability.
The fundamentals of MLOps and how they fit with ML and software engineering, then a hands-on project with ZenML: setting up the environment, designing workflows, cleaning data, experiment tracking, model evaluation, deployment and monitoring, plus a Streamlit app.
Build and ship a complete pipeline: an MLflow server on AWS, a versioned DVC pipeline, model evaluation and registration, a Flask API in Docker, a Chrome extension, and CI/CD deployment on AWS.
A quick start to running LLMs in production: set up a Kubernetes environment and get the vLLM production stack running, the official setup for scalable, high-throughput model serving.