Practice real-time recommendation vocabulary: latency requirements, edge serving, offline pre-computation, online feature stores, and streaming event updates.
0 / 5 completed
1 / 5
'Sub-100ms recommendation _____.' What metric describes how fast a recommendation must be returned?
'Recommendation latency' is the time from request to response. Sub-100ms is the typical target for real-time recommendations to avoid perceptible delay for users.
2 / 5
'The model is served at the _____.' Where is the model deployed to minimize latency?
Serving the recommendation model at the edge (CDN edge nodes close to users) reduces network round-trip time, enabling lower latency recommendations globally.
3 / 5
'We pre-compute recommendations _____.' When are these computations done?
Pre-computing recommendations offline (batch processing) allows instant retrieval at serving time. The trade-off is that offline recommendations may be less fresh than real-time models.
4 / 5
What is an 'online feature store'?
An online feature store (e.g., Feast, Tecton) serves pre-computed, low-latency features to recommendation models at request time — distinguishing it from the offline store used for training.
5 / 5
'Streaming user events update the recommendation model.' What does this approach enable?
Streaming real-time user events (clicks, views, purchases) into the recommendation model enables it to respond to the user's current session context, producing fresher, more relevant recommendations.