Classic Amazon question. They want to see how you handle massive scale with reliability.

I framed the problem around the peak load during Prime Day, maybe 100x normal traffic. Key components: order ingestion service, inventory reservation, payment processing, fulfillment orchestration.

The big insight that I think won it was discussing how to handle inventory overselling. I proposed a two-phase reservation system: soft reservation on add-to-cart (with TTL), hard reservation on checkout. Used DynamoDB with conditional writes for atomic inventory updates.

For the payment side, talked about idempotency and exactly-once processing using SQS FIFO queues. The interviewer liked that I brought up dead letter queues for failed payments.

Spent the last 10 minutes on monitoring and alerting. Discussed circuit breakers between services and automated scaling based on queue depth.

The LP questions were standard. Had good stories ready for "Dive Deep" and "Bias for Action".