Talks about machine learning are full of hype and models that identify cats or unsuccessfully try to differentiate dogs and loaves of bread. Or they are delivered by seasoned PhDs solving problems I've never even heard about at Google scale. In this talk, we'll explore the domain of refactoring software, and the building of a machine-learning model that provides refactoring recommendations to developers, using their own software repositories as a learning corpus. We'll also look at how we took this model to production and the challenges around understanding and operating machine learning in production. From this talk you'll take away: - a deeper insight into what refactoring really is - machine learning applied in a real product, at a small-scale company, by an average developer, under time pressure - some good examples of what not to do when attempting machine learning in production