ML System Design

Please RSVP ahead of time and bring a government-issued ID for check-in.

As developers, we train models and get metrics on a test set — but how does a model actually become a product that millions of people use?
In this talk, I'll walk through the ML system design of a video recommendation application end to end. We'll look at how visual encoders, text encoders, and ranking models fit alongside the infrastructure that makes them useful: vector databases, feature stores, serving layers, and business logic. We'll also cover the gap between offline metrics like precision, mAP, and recall that we optimize during training, and the online metrics like click-through rate and watch time that actually tell you if the product is working. By the end, you should be able to sketch out the high-level design of a system like this yourself and know which pieces to reach for when it's time to build one.

About the Speaker
Mark Robinson, AI Research Scientist at Kostas Research Institute
https://www.linkedin.com/in/mialbro/

Venue Sponsor:
CIC Cambridge
https://cic.com/about/