Several companies have implemented ML platforms to improve the productivity of their data science teams, including by reducing their dependence on engineering teams. In this post, Ernest Chan summarizes what he gleaned from numerous conference talks and blog posts on ML platforms at 11 companies. Chan describes the key components of ML platforms (with which practitioners interface through UIs, config files, and/or APIs/libraries): feature store, workflow orchestration, mode registry, model serving, and model quality monitoring. He also highlights some more unique capabilities, including Intuit’s serverless experience that eliminates the need for users to specify resource configurations; Netflix’s model management platform, which identifies the dependencies between models and features; and PayPal’s inference engine, which enables the flexible composition of models.