Free MLOps workshop April 14

A free online workshop on Tuesday, April 14, is pitching a crash course in how machine learning systems get from notebooks into production. (x.com) Machine learning operations, or MLOps, is the work of packaging, deploying, testing and watching models after launch, not just training them once. DataTalks.Club describes the field as the gap between model development and production deployment, where teams deal with drift, reproducibility and monitoring. (datatalks.club) The workshop agenda points at the tools companies use for that handoff: continuous integration and continuous delivery pipelines on major cloud platforms, Docker containers, Kubernetes clusters, Kubeflow workflows, feature stores, model monitoring and large language model operations, or LLMOps. Docker’s docs say containers package applications, and Kubernetes docs say Kubernetes automates deployment and scaling of those containerized apps. (x.com) (docs.docker.com) (kubernetes.io) Kubeflow is one of the workflow layers on that list. The Kubeflow project says Kubeflow Pipelines is built for portable, scalable machine learning workflows on Kubernetes, which makes it a common way to turn a training script into a repeatable production pipeline. (kubeflow.org) Feature stores solve a narrower problem: keeping training data and live prediction data consistent. Feast, an open-source feature store, says teams use it to define, manage, discover and serve features for both training and inference. (feast.dev) Model monitoring covers the step after deployment, when teams check whether a model is still behaving as expected. Google Cloud’s Vertex AI documentation says monitoring tracks behavior, health and performance, including traffic, latency and errors, so teams can troubleshoot problems in production. (docs.cloud.google.com) The workshop also includes LLMOps, which extends those production practices to systems built on large language models. Microsoft Learn defines LLMOps as the tools and processes for developing, deploying and maintaining large-language-model applications end to end. (learn.microsoft.com) That mix of topics mirrors the way free MLOps training has evolved over the past year, with courses such as DataTalks.Club’s MLOps Zoomcamp bundling experiment tracking, deployment, monitoring, Docker and continuous integration and continuous delivery into one path. Its public repository lists seven sections, from introduction through monitoring and best practices to a final end-to-end project. (github.com) (datatalks.club) For practitioners, the pitch is practical: hands-on labs on April 14 for the stack many teams now expect in production machine learning. The post advertising the session frames it for people building real systems, not just training models. (x.com)