Machine learning is at the core of Pinterest. Pinterest personalizes and ranks 1B+ pins, 700+ million boards for 100M+ users all over the world, using data gathered from collaborative filtering, user curation, web crawling, and more. At Pinterest we model relationships between pins, handle cold-start problems and deal with real-time recommendations. In this presentation Jure gave an overview of the problems and effective solutions developed at Pinterest. He focused on systems and effective engineering choices made to enable productive machine learning development and enable multiple engineers effectively develop, test, and deploy machine-learned models.