Scaling FastAPI from 180 1300 Requests/sec: What Actually Worked

A FastAPI backend stuck at ~180 req/sec with 4s p95 latency was optimized to ~1300 req/sec under 200ms p95 on the same hardware. Key fixes included consolidating 14+ microservices into 4 domain services, switching to truly async drivers (asyncpg, httpx), offloading heavy work to background tasks, resolving N+1 queries and missing indexes, and adding Redis caching. Architecture and database changes delivered ~85% of the gains, while runtime tuning contributed only ~10–15%.