PolicyGuard

PolicyGuard: Neuro-symbolic framework converts organizational policies into executable compliance review engines

PolicyGuard is a neuro-symbolic framework that converts organizational policy documents into typed relational logic rules and atom-level extraction questions, enabling auditable LLM-assisted document compliance review. During review, LLMs answer local questions grounded in retrieved document evidence, while a symbolic evaluator applies formal rules to detect non-compliance. The system is instantiated and evaluated on NDA compliance review against company-specific negotiation policies. The separation of policy formalization, document interpretation, and symbolic evaluation addresses the opacity and maintainability problems of end-to-end LLM prompting for compliance tasks.