As part of my work to help release some COVID-19 models, I wanted a simple workflow that worked on Kubernetes and standard CI/CD services like Gitlab. I ended up with a simple Docker-based multi-stage build to train and then serve the model. This is then pushed to K8s for deployment.
And if you don’t know already, my company, Winder Research, helps organisations design, build and harden data science applications and processes. If you need any help, or if you know anyone that does, please get in touch.