LLMs can replace rigid rule-based systems for document compliance verification, handling messy real-world text better than traditional NER while requiring no task-specific training data.
This paper applies large language models to automatically verify whether securities meet eligibility criteria for use as collateral at the German Central Bank. Instead of manually reading through complex, bilingual prospectuses, the system uses LLMs to extract, normalize, and interpret financial and legal information, achieving 91% precision while avoiding false acceptances.