dense-vectors

BM25 ranks documents by term frequency, which breaks the moment your words don't match the document's words. Here's why dense vector embeddings handle intent queries better, and how TraceMind uses both.