⭐ Design Yelp — System Design Interview Guide

Medium · Geospatial & Reviews

Design a local business discovery platform like Yelp or Google Maps where users can find nearby restaurants, services, and attractions; read/write reviews; and see ratings.

Open the interactive Yelp design on PrepGrind → Drag load balancers, caches, databases, and queues onto a canvas, run a live traffic simulation to watch latency and bottlenecks under load, and follow the full interview walkthrough below — free, in your browser.

Functional requirements

Search businesses by name, category, or keyword near a location
View business details: hours, photos, menu, contact
Write and read reviews with star ratings (1-5)
Filter by rating, distance, price range, open-now
Business owner portal: manage business info
Add/upload photos to business or review

Non-functional requirements & scale

100M businesses worldwide; 500M reviews
Search results < 200ms P95
Proximity search: find businesses within N km
Review write latency < 1 second
Rating aggregation always consistent with stored reviews
Photo CDN: images served < 500ms globally

Capacity estimation

Geospatial search + rich text search combined. Business search = proximity (geospatial) + keyword match + filter. Geohash or S2 cells for proximity indexing. Elasticsearch natively supports geo_point and text search combined. Reviews: write-heavy on popular businesses (rate limiting), read-heavy always.

Core entities

Business — businessId, name, category, location (lat/lng), address, geohash, rating, reviewCount, priceRange, hours, photos[]
Review — reviewId, businessId, userId, rating (1-5), text, photos[], createdAt, helpfulCount
User — userId, username, reviewCount, profilePhoto, elite (bool)

API design

GET /api/v1/search?q=pizza&lat=37.7&lng=-122.4&radius=2km&sort=rating — Search businesses near location.
POST /api/v1/businesses/:id/reviews — Write review. Body: { rating, text, photos? }.
GET /api/v1/businesses/:id — Business details with paginated reviews.

High-level design

Search: Elasticsearch with geo_point + text + filters. Reviews: MySQL with denormalized rating on Business table. Photos: S3 + CDN. Geohash-based proximity index partitions the geographic search space.

Deep dives

📍 Geospatial Search with Geohash

Geohash: encode lat/lng as string prefix. Properties of geohash string: (1) Shared prefix = geographic proximity. (2) Longer prefix = smaller area. "9q8yy" = San Francisco 1×1km cell. Business geohash stored in Elasticsearch geo_point field. Search: filter by geohash prefix = fast proximity search. Edge case: businesses near geohash boundaries → query adjacent cells (8 neighbors).

⭐ Rating Calculation

Naive: SELECT AVG(rating) FROM reviews WHERE businessId = X. Slow on 10K reviews. Better: denormalize — Business table has ratingSum and reviewCount. On new review: UPDATE business SET ratingSum = ratingSum + ?, reviewCount = reviewCount + 1. Rating = ratingSum / reviewCount. Must use database transaction: INSERT review + UPDATE business atomically. Review deletion: subtract from ratingSum.

🛡️ Review Spam Prevention

Yelp has sophisticated anti-gaming: (1) Review gating: new business owner cannot solicit reviews. (2) IP/device fingerprinting: multiple reviews from same source. (3) Review history: new account + only 1 review to competitor = suspicious. (4) ML model: scores review authenticity. (5) "Not currently recommended" section for filtered reviews. Elite users: trusted reviewers with verified history.

🗺️ Maps Integration

Business location displayed on map. Google Maps Embed API or Mapbox for map display. Geocoding: convert address to lat/lng on business creation (Google Geocoding API). Reverse geocoding: "restaurants near me" uses browser geolocation API → lat/lng. Routing: show directions from user to business (Google Maps link or native map app). Map tiles served from CDN.

Scaling considerations

Elasticsearch for combined geo + text search (native support)
MySQL sharded by businessId; reviews partitioned by businessId
Redis cache for popular businesses (top 10K businesses get 80% of traffic)
CDN for all photos — critical for perceived performance
Geohash search: query precision = balance between accuracy and over-fetching

What interviewers expect by level

Junior: Describe business search with location filter, review creation, star rating.
Mid: Geohash proximity search, denormalized rating, Elasticsearch geo_point queries.
Senior: Adjacent geohash cells for boundary cases, anti-spam review system, rating denormalization with transactions.
Staff: Personalized ranking (preferences + history), global CDN for 500M review photos, GDPR for review deletion.

Practice more system design case studies

PrepGrind runs entirely in your browser, free, no installation required. Loading the interactive playground…