As machine learning becomes more pervasive across industries the need to automate the deployment of the required infrastructure becomes even more important. With data velocity increasing every day it becomes more and more important to keep models fresh. Combined with the ever growing popularity of Kubernetes, a full-cycle, containerized method for maintaining model freshness is needed.

In this talk we will present a containerized architecture to handle the lifecycle of an ML model. We will describe our technologies and tools used along with our lessons learned along the way. We will show how fresh training data can be ingested, models can be trained, evaluated, and served in an automated and extensible fashion.

Attendees of this talk will come away with a working knowledge of how a machine learning pipeline can be constructed and managed inside Kubernetes. All code presented will be available on GitHub.