Preparing your design workspace

Loading the problem and your latest saved attempt…

Design a Searchable Emoji Picker

Serve a localized, accessible emoji catalog with responsive search, deterministic selection, and privacy-bounded personalization.

Problem statement

Design a searchable emoji picker for messaging and document clients. For this exercise, assume a versioned canonical catalog contains stable emoji identities, localized names and aliases, categories, variants, and compatibility metadata.

A user can browse categories, search by normalized tokens, choose a supported variant, and emit one stable identity into the host application. The client keeps privacy-bounded per-device recents and favorites; these are never used as global popularity evidence.

Catalog snapshots support offline browsing and search, while an online service publishes signed successor versions and optional aggregate ranking signals. Keyboard navigation, focus order, screen-reader labels, and unsupported-rendering fallback are required.

Define catalog publication, query and selection contracts, localization, ranking, caching, offline behavior, accessibility, privacy, rollout, and recovery. Designing the host editor, keyboard input method, or rendering engine is outside scope.

What to build

  1. Browse a versioned localized catalog by category and supported variant.
  2. Search normalized localized names and aliases with deterministic ranking and fallback.
  3. Return one stable compatible emoji identity to the host under retry-safe selection semantics.
  4. Maintain bounded per-device recents and favorites without leaking private use into global ranking.

At scale

  • For this exercise, assume a 10,000-entry canonical catalog, 10 million daily users, and 20,000 peak search requests per second across online clients.
  • For this exercise, cached search should complete within 100 milliseconds at p95 and an online miss within 300 milliseconds at p95.
  • For this exercise, a published catalog version should reach 99% of online clients within 24 hours while older supported snapshots remain readable.
  • For this exercise, loss of the online service must preserve browsing and search from the last verified snapshot, with visibly stale ranking rather than hidden failure.