© 2021 Strange Loop
Nearly 400 million people in 178 markets around the world open their Spotify app at least once every month and see a brand new set of recommendations from a pool of 70+ million songs and podcasts. Those recommendations on the home page, in your playlist creation screen, from your search bar, and in your Discover Weekly are driven from Machine Learning models powered by a complex set of tools working in unison with the sole purpose of building a more intelligent Spotify product.
In this talk, Josh Baer – product lead for Spotify’s Machine Learning Platform group – will walk you through the lifecycle of a single song recommendation and the complex software that orchestrates the various backend, data and ML specialized libraries built to make the lives of Spotify engineers easier. He’ll describe the challenges of creating a Platform that trains hundreds of models daily and serves millions of predictions per second for 50+ teams building ML for Spotify, and briefly discuss why there are so many companies building similar tooling.
Josh leads a group at Spotify with the aim to build better, more reliable ml-powered products through platformized tooling and infrastructure. He's spent the pandemic hanging out in an apartment in Brooklyn with his partner, two cats, and most recently his baby girl. He holds a MS in CS from NYU and a BS in Philosophy/CS from the University of Pittsburgh.