Semantic Vector Search uses deep learning to associate products with queries in a shared semantic vector space. An encoder model learns from product discovery signals to encode products and queries as vectors. Incoming queries are encoded on the fly, and then products that are “near” the query in the shared vector space are returned.