Design Amazon's Product Recommendation System

Problem

Design a product recommendation system for Amazon.in that serves personalised product suggestions on the homepage, product detail pages, and cart page.

Requirements

Functional:

• Show top-N personalised recommendations per user
• Recommendations must update within 1 hour of a new purchase
• Support "Customers who bought X also bought Y" (item-item)
• Support user-personalised feed (user-item)

Non-functional:

• 500M users, 300M products
• p99 latency < 50ms for serving
• 1M recommendation requests/second at peak

What to cover

1. High-level architecture
2. Offline model training pipeline vs online feature store
3. Candidate generation → Ranking → Serving layers
4. How you handle cold-start for new users/products
5. A/B testing and feedback loop