Towards reducing hallucination in extracting information from financial reports using Large Language Models