BeautyGuard

Abstract:

We propose a stakeholder-informed multi-agent LLMroundtable that mirrors enterprise roles: a LegalInterpreterfor poliey reading, a Rule Checker for item-by-item scanning, a Precedent Researcher for internal/external caseretrieval, and a Risk Planner for graded mitigation. Multi-agent research shows that role decomposition, critiqueand consensus can improve reasoning and planning over single models. Our system has each agent producetraceable, role-specific judgments that are then synthesized into a structured plan, shifting from rejection-based review to proactive risk paths aligned with how cross-functional teams actually work.

For more details please check here .