Problem
Design a nearby restaurant recommendation system that ranks a personalized feed of deliverable restaurants for users opening a food delivery app
- Rank deliverable candidates under a strict 200 ms latency budget while personalizing based on implicit intent
- Optimize a composite objective balancing conversion, ETA, quality, and supply-side fairness to prevent popularity spirals
- Maintain minute-fresh restaurant state features and mitigate cold-start risks for new users and restaurants
Key Test Points
- Two-tower retrieval model
- Wide & Deep ranking network
- Exploration slots with Thompson sampling
Spot an issue or have a suggestion?