Designing better ML systems: Learnings from Netflix

Data science at Netflix goes far beyond eponymous recommendation systems and touches every aspect of the business, from optimizing content delivery to fighting fraud. Netflix’s unique culture affords its data scientists extraordinary freedom of choice in tools, which results in an ever-expanding set of machine learning (ML) approaches and systems. In 2019, Netflix open-sourced Metaflow, its human-centric ML platform. In this session, Netflix shares some lessons learned in its multi-year journey building the ML systems that Metaflow incorporates, covering a diverse range of scale from one-time experimentation on laptops to large-scale model training and serving systems on AWS.