ML Maturity: Unifying and Measuring Processes Across Multiple Teams.

Flo is an AI-driven company. We use machine learning in nearly every part of our product, from cycle predictions to automatic moderation in the community. Our models span years of development, changing architectures, data sources, and frameworks. To support such diverse and long-running tasks, we develop ML infrastructure and common principles that allow us to reduce time to production by up to 80 percent. In this speech, we describe some of our approaches and tools, which address: Computational resources for ML engineers, Unified development environment, Data protection, Model registry, Metric tracking, Experiment reproducibility and other.