Kubeflow is an open-source toolkit for building and operating machine-learning platforms on top of Kubernetes, the system for orchestrating containerized workloads across clusters. It began as an open-sourcing of the way Google ran TensorFlow internally, based on its TensorFlow Extended pipeline, originally as a simpler way to run TensorFlow jobs on Kubernetes, and it has since grown into a broader, framework-agnostic foundation.
Rather than being a single monolithic product, Kubeflow is a modular set of components that cover the stages of the ML lifecycle and can be used together or separately. These include managed notebook environments for development, distributed model training, hyperparameter tuning and neural architecture search, ML pipelines that define multi-step workflows, and model serving. Because everything runs on Kubernetes, teams get its scaling, scheduling, and portability across cloud and on-premises infrastructure for free. The project is now community-led, governed by working groups under a steering committee rather than by a single vendor.
Why a business reader should care: Kubeflow is a common way organizations that have standardized on Kubernetes build a shared, repeatable machine-learning platform, so data scientists can train and deploy models on the same infrastructure the rest of the company already runs.