High-Level Design of the Typeahead Suggestion System

High-level design

Separate interactive prefix lookup from collecting evidence for future rankings. A keystroke request reads a prepared index; recording a completed search or selection can happen asynchronously. This prevents a ranking rebuild from blocking the person typing.

The proposed system should perform two main functions:

• Return eligible ranked completions for the current prefix, with optional personalization.

• Collect approved search events, aggregate popularity, and publish updated candidate data.

In the baseline, each keystroke triggers a prefix request. A suggestion service returns up to ten candidates from Redis or a serving index. Debouncing can reduce requests later. The assembler aggregates search events and builds index versions independently; freshness is bounded by its collection and publication delay.