Design Search Ranking System — System Design Practice | AmanAI Lab

Sheet›Design Search Ranking System

GoogleAmazonMSFTMeta

45:00

Hard

The Problem

Problem

Design a search ranking system for an e-commerce platform with 100 million products and 10,000 search queries per second.

Requirements:

•Return ranked results in ≤ 200ms
•Ranking should maximise purchase conversion rate
•Support personalisation (different rankings for different users)
•Support real-time inventory and price updates
•Enable A/B testing of ranking algorithms

What you'll be assessed on

Multi-stage retrieval and ranking architecture, feature engineering for ranking, learning-to-rank approaches, and how you balance relevance vs business metrics.

Scale & Constraints

▸100M products, 10K QPS
▸Serving latency ≤ 200ms P99
▸Inventory updates reflected in ≤ 1 minute
▸Price changes reflected in ≤ 5 minutes
▸New products indexed within 10 minutes of upload
▸Personalised results for 500M registered users

Must Cover0/10

Hints (if stuck)

💡 BM25 handles keyword and brand name queries; dense retrieval handles "comfortable running shoes."

💡 L1 ranker processes all 1000 candidates; L2 ranker only processes top 100 (expensive features).

💡 Position bias in click logs: correct with inverse propensity weighting before training.

💡 Cache query embeddings (hot queries repeat) — 70% of e-commerce queries repeat within a day.

💡 Price and inventory filters can be applied at the retrieval stage to avoid ranking unavailable items.

0 words · auto-saved

Back to sheet

Mock interviewTake quizFlashcards

Problem

Design a search ranking system for an e-commerce platform with 100 million products and 10,000 search queries per second.

Requirements:

•Return ranked results in ≤ 200ms

•Ranking should maximise purchase conversion rate

•Support personalisation (different rankings for different users)

•Support real-time inventory and price updates

•Enable A/B testing of ranking algorithms

What you'll be assessed on