A specialist that speaks Polish fluently and does one thing: deciding which documents belong at the top of a search result. It takes a query and a passage, then outputs a relevance score rather than generated text. Its 514-token context window is tight, so it works best with concise passages rather than long documents.