Vector Search at Uber - Implementing and Scaling KNN Search with Apache Lucene

Explore the implementation and impact of Vector Search at Uber in this 24-minute conference talk. Dive into the revolution of Search and Information Retrieval since the BERT and GPT breakthroughs in 2018. Learn how Uber incorporated Vector Search support using Apache Lucene for their Search platform, enabling critical use cases like Semantic Search and Gen AI support. Discover the process of integrating KNN search capability with the HNSW algorithm from Lucene 9 and extending it with live vector update ingestion. Gain insights into the learnings and optimizations required to scale the vector search solution to meet Uber's demanding requirements. Examine a real-world application through the collaboration between Uber's Search and Maps teams, showcasing how Geo-Semantics search significantly improved search quality by vectorizing semantic signals. Presented by Yupeng Fu, Principal Engineer at Uber, who leads the Real-time Data and Search Platform team.

Vector Search at Uber