Born in Bradford Research Assistant: A Deployed Retrieval-Augmented Generation System for Health Data Discovery

Large longitudinal health studies distribute information across various sources such as data-dictionary variables, questionnaires, collection documentation, and publications, making research data discovery time-consuming. We present the Born in Bradford (BiB) Research Assistant, a deployed retrieval-augmented generation (RAG) system for navigating BiB metadata and research literature. The system combines exact entity lookup, hybrid retrieval, reranking, and evidence-linked generation, and is currently being piloted by BiB team members. We describe its integration with the existing data-dictionary workflow, and report retrieval evaluation on 300 grounded queries, and response evaluation on 105 test queries.