A specialist that converts Korean text into dense vector embeddings using a contrastive learning approach (SimCSE). It operates with a very small context window of 514 tokens, so it works best on shorter passages. Its output is purely numerical — no generated text, just representations of meaning.