Bloomberg System Design Interview: Design Search Typeahead

That phrasing is close to verbatim from a candidate account. Instead of "design Google search suggestions", it arrives as two function signatures, which tells you where the hour is going to be spent. Another account calls it a lengthy problem, meaning the structure work fills the time rather than the architecture diagram.

Treat it accordingly. Most system design preparation would have you draw boxes for ten minutes and then discuss sharding. Here you should get to the data structure quickly, get it right, and let the scale conversation follow from its properties.

Assume these out loud. A corpus of around ten million distinct terms. Roughly a hundred thousand keystrokes per second at peak across all users. A few thousand new terms entering the corpus each second. The keystroke number is the one that matters most, and the next paragraph explains why.

Every keystroke is a request. Someone typing an eight-character query fires eight lookups, not one, so your read rate is roughly eight times your query rate. Doing that multiplication aloud early reframes the whole problem as a read-optimisation exercise, and it justifies precomputation that would otherwise look excessive.

What's inside

1 diagram
15 sections
about 34 min

Premium lessonBloomberg interview guide

These took digging.

What candidates walked out with, from 20,000+ reports, forum threads behind logins and write-ups in Chinese. Then solved step by step.

100 companies. Checked every month.

Unlock all 100 companies

$49 a year · less than $1 a week · Cancel anytime

See how we find them

These took digging.

Know what Bloomberg asks before you walk in.