Book Price Aggregator Platform

Design a Book Price Aggregator Platform that lets users search for a book and instantly see the lowest price available from 50–200 third-party sellers. The system must accept a book identifier (ISBN, title, etc.) from a web or mobile client, concurrently query every registered seller’s HTTP API, and return the best price within a hard deadline of 10–20 seconds. Sellers have heterogeneous APIs (JSON, XML, custom auth), unpredictable latency (100 ms–15 s), and varying reliability (some fail 5–30 % of the time). The platform should therefore implement a parallel fan-out with per-seller timeouts, return partial results (“checked 47 of 52 sellers”), and gracefully degrade when sellers are unreachable. To stay within latency targets you must add a short-lived cache (TTL 5–30 min) that serves repeated queries immediately while asynchronously refreshing prices in the background. Track seller health (response time, error rate, price accuracy) and use these metrics to rank sellers: query the most reliable ones first and optionally issue hedged requests (second call after p50 latency) for critical sellers. The service must scale to 10 000 QPS, tolerate AZ failures, and keep per-query p99 latency < 20 s. Provide high-level design, API contracts, data models, caching strategy, concurrency model, reliability mechanisms, and how you would continuously rank and deprioritize slow or flaky sellers.