BlockEdge

Technology and digital life

Uber Eats search overhaul cuts end-to-end latency by 50%, Uber reports

Screenshot

Originally reported November 16, 2026.

Uber has reworked major parts of the Uber Eats search pipeline, reporting a 50% reduction in end-to-end search latency. The changes span retrieval, feature hydration, ranking, ads, rendering and infrastructure. Uber also introduced an agentic coding workflow to identify, benchmark and validate further optimizations.

The team shifted its focus from backend API response time to above-the-fold completion: the time from a request until the first screen of results, including images, has fully rendered. Server-side caching for pagination shortened the initial response, while asynchronous rendering allowed result items to be processed concurrently. Uber reported an improvement of more than 200 milliseconds in above-the-fold latency.

Other changes reduced work at several stages. Uber found that feature data was being hydrated for tens of thousands of candidates before ranking, even though many would be discarded. Removing low-value retrieval strategies saved about 120 milliseconds; product-level embeddings cut data queries by more than 100-fold and saved another 50 milliseconds. Separating ranking features from display data reduced latency by more than 100 milliseconds. Uber also attributed reductions of 35 milliseconds and 40 milliseconds to removing unnecessary dependencies and request hedging, respectively. A redesigned ads path, using column-oriented bidding data, in-memory access and less serialization, cut about 130 milliseconds. These are reported improvements to different parts of the pipeline, not a single additive latency figure.

Uber is exploring, rather than announcing as completed, end-to-end microbatching, product-based retrieval, Zero Pass Ranking and HTTP multipart streaming. It reported that early tests of product-based retrieval reduced p99 latency by more than 50%.

Source: infoq.cn (originally published 2026-11-16 17:24 UTC+8)