Approach: Use multi-layered architecture with load balancing, message queues, caching for hot products, and horizontal scaling to handle 10x traffic spikes during peak events.

Considerations: Balance consistency vs. availability with eventual consistency for inventory; trade real-time accuracy for high throughput, especially for seller counts and low-stock items.

Key Points: Implement Redis caching for trending items, Kafka for async queues, database sharding by region/category, circuit breakers for payments, and graceful degradation.

Approach: Design distributed system with event-driven architecture where sellers update inventory, reservations lock stock temporarily, and reconciliation syncs across regions.

Considerations: Handle replication lag across Southeast Asian data centers; balance seller update frequency and system load to avoid inconsistency and lost sales.

Key Points: Use event sourcing for audits, reservation timeouts (2-5 min) to free inventory, region-aware replication with conflict resolution, and real-time fraud detection dashboard.

Approach: Implement multi-stage ranking pipeline with Elasticsearch and BM25 ranking, apply learning-to-rank models trained on user behavior, and personalize by region, language, device.

Considerations: Optimize for mobile-first usage; balance personalization vs. cold-start for new sellers; handle language diversity affecting search relevance.

Key Points: Use Elasticsearch per region, train collaborative filtering on buyer-seller data, apply real-time A/B testing, implement fast inference for 100ms targets, handle multi-language queries.