Rating is available when the video has been rented.
This feature is not available right now. Please try again later.
Published on Jul 24, 2019
Delivering a data science to production currently looks something like this: Clara your brilliant data scientist creates a model in her notebook that can have a huge impact on the companies revenue after working on it isolated for a few weeks a pickle file is born and Clara has to find a way to expose to users, with any luck she will grab the correct product manager that will be able to maybe prioritize it during the next sprint, once the engineering team gets to it the model is already stale and the process resets. This familiar scenario wastes resources from multiple departments and introduces overhead instead of value. In this talk I want to show how to remove a major bottleneck between data science and production. Assaf a generic model serving framework and a few devops methods you can free your scientist to do science and move quickly while decoupling them from engineering efforts, allowing engineers to deliver value, stop doing repetitive work and focus on what they are best at. Assaf abstracts away caching and networking allowing scientists to use it as a library in their favorite language and focus on what's important to them. Assaf relies on persistent event streams to communicate between users and services and come with built-in metrics collection and tracing so deployed models are resilient and can be monitored efficiently, this highlights hotspots and triggers a darwinian process making teams focus on what's they know is important and reduce guessing to a minimum.