Internal policies had to stay aligned with a large and constantly changing body of central bank regulation: anti-money-laundering notices, circulars and historical rulings. Manual reviews were slow and expensive, and gaps were easy to miss.
What we built- An in-house large language model, fine-tuned on central bank regulatory data and running on the bank's own servers.
- A structured knowledge base of roughly 25,000 to 30,000 historical regulations and notices, with vector-embedded comparison.
- Each internal policy is checked against the closely related requirements to return a compliance score, flag contradictions and gaps, and recommend improvements.
- Historical rulings are cross-referenced so every assessment is grounded in precedent.
- Manual policy review time reduced by 72%.
- 10% of policies found to have compliance gaps that manual review had missed.
- Data never leaves the bank: the model and knowledge base run on-premise.