This document discusses KubeFlow and data science in Kubernetes. It provides an overview of the KubeFlow project, which aims to make machine learning (ML) workflows portable and scalable on Kubernetes. Some key KubeFlow components discussed are JupyterHub, Katib for hyperparameter tuning, Kubeflow Pipelines for workflows, and serving components like KFServing. Lessons learned include that KubeFlow is well-suited for scaling deep learning but favors Google Kubernetes Engine, and that data and notebook versioning could be improved.