A compact, fast embedding specialist tuned specifically for question-answer retrieval tasks. It excels at matching questions to relevant passages or documents, making it reliable for semantic search pipelines. Its small size (6 layers) means it trades some representational depth for speed and low memory footprint.