HR-Agents: Using Multiple LLM-based Agents to Improve Q&A about Brazilian Labor Legislation

A collaborative team of 17 Brazilian researchers has introduced 'HR-Agents,' a novel multi-agent system designed to tackle the intricate challenge of answering questions about Brazil's Consolidation of Labor Laws (CLT). The system, built using the CrewAI framework, moves beyond a standard single Large Language Model (LLM) approach by deploying multiple specialized AI agents that work cooperatively. Each agent handles distinct aspects of employment law, and the architecture integrates Retrieval-Augmented Generation (RAG) to ground responses in relevant legal context, aiming to reduce misinformation and improve reliability for Human Resources professionals.

The research, presented at the XVII Simpósio Brasileiro de Automação Inteligente (SBAI) in July 2025, rigorously evaluated the system against a baseline RAG pipeline powered by a single LLM. The team used automated metrics like BLEU, LLM-as-judge evaluations, and crucially, assessments by human legal experts. The results demonstrated that the multi-agent approach significantly improved both the coherence and factual correctness of responses compared to the simpler baseline. This validates the potential of coordinated AI agents to manage specialized, high-stakes domains where accuracy is paramount.

By demonstrating a functional architecture for AI-driven legal assistance, this work provides a blueprint for applying multi-agent LLM systems to other complex regulatory and compliance domains. It directly addresses the inefficiencies and delays inherent in traditional methods of legal inquiry, offering a scalable path to streamline HR operations and enhance labor law compliance through more accurate and accessible information retrieval.