Agentic RAG for Structured SFIA Skill and Level Extraction

A three-agent retriever-matcher-verifier framework outperforms lexical and single-agent RAG methods for extracting professional skills and responsibility levels from text. This approach maps free text to the SFIA framework's closed vocabulary of 147 skills and seven levels.