Not Just Buttons: Feature Flagging Your Machine Learning Architectures

Feature flagging is common practice in the UI and front end layers of modern web-based applications. The ability of feature flags to allow selective roll-out of new UX opens many more doors in the world of product possibility - and the benefit of using a service like Launch Darkly to enable this is easy to see. At Demandbase, while we initially used feature flagging to control what our customers see in our platform, we soon saw the potential for a similar paradigm to impact backend APIs and even our machine learning modeling itself. In this talk, you’ll learn a bit more of our history with feature flagging and LaunchDarkly, and how this led us to explore using feature flags to control production machine learning workflows on Spark. In doing so, we were able to experiment faster - accelerating our learnings in a real production environment. More than just supporting R&D though, by bringing feature flagging into our machine learning workflows, we are able to deliver more value more quickly to our customers, in a safer and more robust way than ever before.