A city-scale ride-hailing platform was carrying rides, drivers and payouts on a backend that could stall as a single unit: one blocked event loop took health checks, dispatch and booking down together, and the analytics everyone needed ran against the same live tables the riders did.
Rebuilt the core as a two-worker FastAPI service with a circuit breaker on every hot Redis path, cluster-safe queueing with a fallback queue behind it, and reporting moved off live tables onto materialised views refreshed on a schedule — revenue by day, performance by provider, demand by hour of the city.
Customer app, provider app, vendor console and web run against one hardened core in production on AWS. A stalled dependency now degrades one path instead of the platform, and a question about last month's demand is answered in seconds rather than avoided.
Want a system like this?
Tell us what you're building. We'll respond with a scoped plan, not a sales pitch.