Orrin Labs' 1,536-dimension embedding model for retrieval, semantic search, and classification, with text and image inputs.