“The first version of Glean's product used transformer models for semantic matching, a technique they internally called "embedding search," before the term "vector search" was popularized.”