Technologists often opine that software engineering standards and best practices should be applied when developing machine learning systems. However, they rarely clarify how these standards and best practices can be applied when building data-defined products. In this post, Laszlo Sragner describes the systems and processes that engineers adapted to enable agile software development (e.g. testing, decoupling, refactoring) and discusses how these may be used by ML engineers while exploring data, implementing a POC and MVP, and during and after productionization