#two-tower

The architecture behind every recommendation system that actually works at scale — why splitting the model in half is the key to serving billions of candidates in milliseconds.