Proximity Service for Nearby Businesses

Proximity Service for Nearby Businesses is a system design interview question commonly asked by Amazon, focusing on backend, geolocation, web services, and scalability with machine learning elements for ranking.[1]

Problem Statement

Design a proximity server that stores and reports distances to places like restaurants, enabling users to search for nearby businesses by distance, popularity, or other metrics. The system must handle data for hundreds of millions of businesses globally, supporting features like business CRUD operations (add, update, delete) by owners and efficient user queries for locations within a specified radius.[2][1]

Functional Requirements

• Users query nearby businesses given latitude/longitude and optional search radius (e.g., return all within 500m).
• Support sorting by distance, rating, or popularity.
• Business owners can create, update, or delete business listings (name, location, category).
• High availability for read-heavy traffic with spikes (e.g., 100M DAU, 5K QPS).[1][2]

Input/Output Examples

No formal coding-style input/output formats exist, as this is a system design question. Instead, APIs are typically outlined:

Search API (GET /nearby)

• Input: { lat: 37.7749, lng: -122.4194, radius: 500, category: "restaurant", sort: "distance" }
• Output: [{ business_id: 123, name: "Italian Bistro", lat: 37.7750, lng: -122.4180, distance: 120m, rating: 4.5 }, ...] (top results within radius).[2]

Business Management API (POST /business)

• Input: { name: "New Cafe", lat: 37.7760, lng: -122.4200, category: "cafe" }
• Output: { business_id: 456, status: "created" }.[2]

Constraints and Scale

• Data: 200M+ global businesses; location data ~5GB (lat/lng per business at 24 bytes each).
• Traffic: 100M DAU, ~5K QPS peak; read-heavy (no writes during queries).
• Performance: Low-latency proximity searches (<100ms); handle 2D geospatial queries efficiently.
• Challenges: Avoid full table scans; use geospatial indexing (e.g., geohash for grid-based queries covering 9 neighboring cells); no real-time friend tracking unless specified.[1][2]