Let me take you onto an in-depth journey of the challenges and unique solutions it took to deliver web-scale vector search in the e-commerce domain. Large-scale shopping platforms come with specific challenges that have traditionally been an issue for vector-search. Those challenges include – among others – millions of updates per day because products go in and out of stock or prices change, large-scale searches with advanced filters, such as price ranges while keeping strict latency requirements, and last but not least the challenge of matching exact keywords. All three have traditionally been very hard for embedding-based search and many vector databases still struggle with those today. 

In this tech talk, we will walk you through the unique solution it took to support web-scale e-commerce users including novel freshness techniques on graph-based algorithms, use of roaring bitmaps for range searches, and Hybrid Search.