Amazon System Design Interview Questions

This is a relational problem with high transaction volume and read-heavy queries. Use normalized tables:

• Customers (customer_id, name, contact info)
• Orders (order_id, customer_id, order_date, status)
• Order_Items (order_id, product_id, quantity, price)
• Products (product_id, name, category, inventory_count)

Index frequently queried fields like product_id and customer_id. For large-scale reads (e.g., recommendations), replicate to read-optimized stores or search indexes.

E-commerce requires product discovery, cart management, payment, and fulfillment integration. Use a microservices architecture:

• Catalog Service for product data
• Search Service with Elasticsearch/Solr
• Cart Service with session or user-based storage
• Checkout/Payment Service integrated with fraud detection
• Order Management tied to inventory and shipping
• Scalability comes from independent scaling per service and aggressive caching of product data.

A returns chat system must be real-time, secure, and context-aware. Store conversations with a reference to the related order/return_id. Use WebSockets for low-latency messaging and integrate automated bots for FAQs or return policy guidance. Agents should see the customer’s order history and previous chats instantly. All messages must be logged for compliance and support quality review.

The system must manage locker availability, reservations, and secure access codes. When a package arrives, the system assigns it to an available locker, updates its status, and sends a pickup code to the customer. At pickup, the locker verifies the code and updates the order to “delivered.” The backend must integrate with logistics systems and sync locker inventory in real time to prevent over-allocation.

Use a route optimization engine that factors in traffic, distance, driver location, and delivery deadlines. The system ingests orders, assigns them to drivers using a nearest-driver algorithm, then calculates the best route (possibly batching multiple deliveries). Real-time GPS tracking allows dynamic re-routing. Store route history for future optimization and ML model training.

Amazon’s recommendations combine collaborative filtering, content-based filtering, and personalized ranking. Batch jobs build long-term preference models, while real-time signals (recent views, cart additions) adjust rankings instantly. The system must handle billions of items and hundreds of millions of users, so it uses a distributed feature store, scalable ranking services, and A/B testing frameworks. Recommendations are context-aware—different for home page, product page, and checkout.