Design real-time top-K POI retrieval on maps

Q: Design real-time top-K POI retrieval on maps

This is a ML System Design interview question from Snapchat for Machine Learning Engineer roles. View the full question and solution on PracHub.

Q: How do I approach ML System Design interview questions?

ML System Design questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master ml system design interviews.

Question

Real-Time Top-K POIs in Viewport: System Design

Context

Design a real-time system for a mobile map that continuously shows the top-K points of interest (POIs) within the user’s current viewport as they pan and zoom. The system must feel instant, remain stable while the viewport moves, and support personalization.

Requirements

Functional

Return top-K POIs for the current viewport rectangle and zoom level, with stable ordering during small pans/zooms.
Support filters (e.g., category, open-now), and pagination/infinite scroll for lists.
Deduplicate POIs across tiles and avoid flicker when moving.
Personalize ordering per user when available; provide sensible defaults for cold start.

Latency/SLOs

End-to-end p95 latency < 150 ms (from gesture end to POIs rendered), p99 < 250 ms.
Freshness: most POI attribute updates visible within 1–5 minutes; critical updates (closures) < 30 seconds.
Availability ≥ 99.9%.

Specify

Architecture across client, edge, and backend services.
Spatial indexing and tiling strategy (e.g., quadtree/S2, R-tree), cell covering, and range queries.
Candidate generation and cell unions.
Retrieval vs. re-ranking pipeline.
Caching (client/edge): tile-level and viewport-level.
Freshness and incremental updates.
Metrics (latency, quality, stability, freshness, cost).
When to use approximate nearest neighbor (ANN) for retrieval vs. learned models for ranking; decision criteria and hybrid approach.
Load shedding, pagination while panning, dedup across tiles, cold start, offline vs. online evaluation, and cost considerations.