Designing an AR Lens Recommender With Value-Weighted Labels

Reported in interviews at Snap

Problem statement

How would you rank augmented-reality lenses when someone opens a social camera app? The catalog is huge, new lenses arrive constantly, and the carousel must load almost instantly. Ranking by tries alone rewards gimmicks, while device compatibility, content policy and negative feedback also matter.

Explain the objective, cold start, candidate retrieval, ranking, serving and fallbacks.

Show hint

Start from what a valuable lens session looks like for the user, and which logged actions are evidence of it beyond a single try.

Relevant roles: ML Engineer, Applied Scientist and AI Engineer.

Related free problems