“Glean builds and pre-trains its own proprietary small language models on each customer's data specifically for semantic matching and creating custom embeddings.”