A faster retrieval approach that converts queries and passages into numerical vectors separately, then compares them for similarity, rather than processing them together.