World Cup Prediction Platform — Built for the Match-Day Spike

A World Cup prediction game has a brutal traffic pattern: thousands of fans sit idle for days, then hammer the app all at once the moment a match kicks off and again at the final whistle. A no-code prototype couldn't survive that spike — and re-computing live leaderboards for everyone in the same second melts a naive backend.

We re-platformed the game onto a self-hosted, horizontally-scalable stack engineered specifically for the match-day surge. The web tier is fully stateless and runs multiple load-balanced replicas; live score and leaderboard updates are pushed over Server-Sent Events and fanned out across all replicas through Redis pub/sub, so one update reaches every connected fan instantly. The expensive hot paths are defended by short-lived caches that turn a thundering herd of identical requests into a handful of database reads, leaderboards are computed in a single indexed SQL query instead of thousands, post-match scoring runs in lock-free batches, and per-user rate limiting (shared across replicas via Redis) stops any one client from overwhelming the system. The whole thing is load-tested with a k6 "big match" scenario that simulates everyone online at once, and a built-in metrics layer watches request rates, latency and event-loop lag in real time.

The platform is engineered and load-tested to keep live leaderboards and predictions fast for tens of thousands of fans hitting it simultaneously during a match — on modest, self-hosted infrastructure, with no per-user cloud bill. Adding capacity is as simple as running more replicas.