DevOps, Security & Observability in ML

Piecing together your stack from the MLOps Jungle

Luke Marsden, MLOps Consulting

Abstract

In this talk we'll look at how you can build:

  • MLOps processes
  • on top of tooling stacks
  • on top of infrastructure

For the MLOps processes, we'll look at the interplay between data, code, models, deployments, and monitoring. We'll look at how models connect to apps to provide business value, and feed back into training data & code changes.

For the tooling stacks, we'll survey some of the popular tools, and look at how they can be wired together on Kubernetes with OIDC. For the infrastructure, we'll look at clusters, object storage, git, GitOps with ArgoCD and using Keycloak as an OIDC provider.

We'll talk about observability (e.g. statistical monitoring of models), and ontology (what is there). We'll review the MLOps principles of reproducibility, accountability/lineage, continuous delivery and collaboration.

And I'll give a sneak peek of something totally new we're working on.

Bio

Luke is a founder of the MLOps Community with Demetrios. He is a bit of an entrepreneur and was involved in the early Docker and Kubernetes days and building an end-to-end MLOps platform. He now hacks on various projects in the MLOps space, including with David Aronchick, the founder of Kubeflow.

Report abuse