A compact Russian-language retrieval specialist built on the RuBERT family. It encodes queries and passages into separate vector representations, making it suited for semantic search pipelines where query and document embeddings are computed independently. The 512-token context window is a practical constraint to keep in mind for longer documents.