Design a Game Recommender from Retrieval to Ranking
Company: Roblox
Role: Machine Learning Engineer
Category: Machine Learning
Difficulty: medium
Interview Round: Onsite
Design a game recommendation system, explaining both candidate retrieval and ranking in detail.
### Constraints and Clarifying Questions
The report does not specify labels, traffic, eligibility rules, or a serving objective. Give a complete illustrative baseline and label its choices. Clarify what counts as a useful recommendation, how newly released games enter the system, and which user or game constraints must be enforced.
### What a Strong Answer Covers
A defined prediction target, timestamped data, retrieval training, a concrete ranker and loss, a serving score, cold start, and offline/online evaluation aligned with the goal.
### Follow-up Questions
- How would you avoid recommending only already-popular games?
- How would a newly released game receive candidates before it has play history?
Overview: Design a game recommendation baseline with a qualified-play target, two-tower retrieval, and probability ranking. Cover new games, point-in-time labels, feedback bias, compatible serving versions, and aligned evaluation.
Read the full Roblox Machine Learning Engineer interview experience this question came from