I can't and shouldn't spill the beans on how Uber's matching algorithm actually works — mostly because the beans are under an NDA and also very well caffeinated. What I can say: it's not the mystical black box described in many system-design blogs. The closest answer I got was from Gemini, but we have evolved from that years ago added more sophisticated sharding and routing logic and custom features specific marketplace and added forecasting and prediction models to improve matching. On lighter note, I wonder what would be latency and throughput if our matching algorithm recommended matchmaking by astrology. Spoiler: we evolved past zodiac-based sharding.