Grounding large language models at runtime involves querying databases or knowledge graphs to retrieve factual evidence from systems of record