This is something that a small LLM handles quite well. I’m using one to normalize MTA delay reports so I can aggregate stuff in a sane way.