Recommender Systems Engineering
Millions of items. One ranked feed.
Candidate generation, two-tower retrieval, feature crosses, bandit exploration, and the feedback loops that make recommenders eat their own tails — the full production stack.
Module
The Two-Stage Funnel
Why every production recommender retrieves hundreds cheaply before it ranks dozens expensively
Module
The Ranking Stack
Two-tower models, feature crosses that memorize, and the business layer that gets the last word
Module
The Feedback Problem
The system generates the data it learns from — loops, bubbles, bandits, and metrics that lie politely
Module
Cold Start & the Serving Layer
New users, new items, new catalogs — and the physics of freshness
Capstone project
Build a recommender
Build a small recommender on a real dataset (MovieLens or similar). Compare a popularity baseline against a collaborative-filtering approach and measure the lift.
Deliverable: A notebook with both approaches, a metric comparison, and where the smart version wins.
Milestones
Finish every lesson to earn the Recommender Systems Engineering Certificate.