Blogs›Uber›Delivery Search Platform Evolution

Delivery Search Platform Evolution

posts

2026

Uber has significantly evolved its Delivery Search Platform, moving from traditional lexical matching to a sophisticated semantic search system. This evolution addresses challenges in understanding complex user queries, including synonyms, typos, language mixes, and context. The platform now leverages deep learning models (Qwen backbone) with a two-tower architecture for real-time query embedding and batch document embedding. Key advancements include optimizing retrieval accuracy against infrastructure and compute costs through careful tuning of ANN parameters, quantization, and embedding dimensions, and ensuring safe, automated, and rollback-capable index deployments with strong validation, versioning, and real-time consistency checks in production.

2026

Evolution and Scale of Uber’s Delivery Search Platform

4/14/2026

This post details the evolution of Uber's Delivery Search Platform to a semantic search system using deep learning models (Qwen backbone) with a two-tower architecture. It covers model training with PyTorch and DeepSpeed, offline inference using feature stores, and optimization strategies for quality and cost (ANN parameters, quantization, embedding dimensions via MRL). It also describes productionization and reliability through a bi-weekly blue/green index deployment strategy with automated validation checks (completeness, backward-compatibility, correctness).