Design Yelp

A local search app answers one question many times a second: what good places are near here? People search where they stand, or pan the map; they filter by category, price and whether a place is open; they read reviews and look at photos; and they write reviews of their own.

Places rarely move, so this is a read problem, unlike dispatching drivers whose positions change every few seconds. The hard parts are finding what is nearby without reading everything, coping with cities packed thousands of places deep and countryside with none, and keeping ratings honest as reviews pour in.

The practice brief states the requirements up front; in an interview, you would confirm them in the first few minutes. Two carry the design: a search returns quickly at a high rate, and a place just outside where you looked still has to be found.

Work out only the numbers this design depends on, with round figures and one line of working each. The data a search needs turns out to be small.

Keep a place’s rating and review count on the place itself, so a search result needs nothing but the place. Give it a status, so a place that closed for good can be left out, and store its hours as a weekly schedule with special days.

Reviews carry a version that rises on each edit, which later lets the rating count each version once. They live in their own store, partitioned by place and month so no famous place makes one enormous partition, with a stored list of each place’s most useful reviews for the other sort.

Search takes either a point with a radius, for near me, or a map box, for the map, plus words, category, filters and a cursor. The cursor holds a point-in-time id and the last result’s ranking value and place id, so scrolling never repeats or skips a place while the index changes (Elasticsearch docs (opens in a new tab)).

Writing a review carries an Idempotency-Key, so a phone that times out and retries doesn’t post twice. Owners edit their place only once verified as its owner. Searching needs no account; writing does.

Build the design one flow at a time, following the brief: a search near me, a pan of the map, and a photo from upload to screen. The deep dives then harden it.