More and more companies than ever before are trying to build and deploy machine learning models to production, but according to research many of those projects fail. WHy? Building ML systems involves dealing with an entirely new type of artifact - the model, which requires a complex infrastructure to be set up around it to monitor its training and inference performance. In response to this need, 100s of companies have sprung up offering various tools and techniques to help make building models and managing them in production easier? But do they fulfill their promise? In this session, we will deep dive into how building ML systems from software engineering, the special challenges that we encounter due to large data sizes, long feedback cycles, non-deterministic behaviour and the frequent need to use GPUs to accelerate our workloads with a specific case study of how systems have been built for a large scale recommender systems at Storytel. Check out our new channel: NDC Clips: @ndcclips Check out more of our featured speakers and talks at https://ndcconferences.com/ https://ndcoslo.com/