Katkat
Back

Design Real-Time Typeahead for Enterprise Search: History, People, Docs, Apps, Filters

GleanFrontend Engineer · Onsite
MediumFrontend EngineerOnsiteSystem Design

Last updated

Design real-time typeahead for an enterprise search product that suggests past searches, people, documents, apps and filters as the user types. Tests client request policy and stale-response handling, accessible combobox UI, permission-aware backend fan-out, blending across sources and latency budgets.

In the system design round of a frontend engineer loop, the question was: how would you implement the typeahead search of the company's enterprise search product? As the user types into the search box, the product shows real-time suggestions drawn from several kinds of source:

  • the user's search history,
  • members of the user's organization (people),
  • related documents,
  • apps,
  • filters.

Design the client and the service behind it.

Clarifying Questions

  • Is the focus the client architecture, the backend, or both?
  • What latency must suggestions meet, and on which devices and networks?
  • How many suggestions are shown, and are they grouped by type or blended into one ranked list?
  • What happens when the user picks each kind of suggestion: run a search, open a document, open a profile, launch an app, apply a filter?
  • How large are the customer organizations, in people and documents per customer?

Hints

Hint 1

Latency is per keystroke: Count how many requests one search can generate as the user types, and decide which suggestions can be answered without any network call.

Hint 2

Not everyone may see every document: Think about what a suggestion reveals even before anyone clicks it.

Follow-up questions

  • How would you personalize the ranking of people and documents for each user?
  • How do you handle a source that is down or slow without blocking the others?
  • How would you support typing in languages that use an input method editor?