A slow API that was blocking user growth, made 10x faster.

Cuez runs live broadcast production, where a three second response is not slow, it is broken. Growth was being held back by an API that got worse as usage rose, and the assumption in the room was that more servers would fix it.

The first pass was measurement, not optimisation. Most of the latency turned out to sit in a handful of query patterns and repeated work that could be done once and cached, not in the parts anyone suspected.

Heavy work moved off the request path into queues with Laravel Horizon, hot paths got proper caching in Redis, and media processing was reworked so it stopped competing with user traffic.

Responses went from three seconds to three hundred milliseconds, a 10x improvement, and the infrastructure bill fell by roughly 40% because the servers were no longer compensating for the code.