Uber Eats Halved Search Time Without Any Big Fix

Uber Engineering detailed how it halved Uber Eats search latency: no single change saved more than about 200 ms, but combined they cut the wait in half. Key moves included an above-the-fold metric, deleting redundant retrieval strategies, splitting hydration and hedging tail latency.

• Above-the-fold metric tracks time until the first screen renders with images
• Removing redundant retrieval strategies saved about 120 ms
• Grouping the catalog by product shrank it roughly 100x
• Request hedging across shards saved another 40 ms