← All posts

Maat: a competition-law research agent with RAG and official citations

A ReAct agent with tools beats general chatbots on merger/case tasks—RAG, web fallback, and clarifying questions.

Tags
AIresearch
Maat: a competition-law research agent with RAG and official citations
Contents

In brief

Maat is a ReAct agent for competition law research—grounding cases in official sources instead of hallucinating precedents like general assistants.

What they studied

Co-designed with experts: RAG, rich citations, web search for coverage gaps, clarification on ambiguous queries. Baselines include Claude, ChatGPT, SaulLM-7B, LegalGPT.

Key findings

  • Significantly better on case-specific tasks.
  • Within range of the top baseline on theoretical questions.
  • Dataset on GitHub.

What this means for developers

Vertical agents need tool orchestration, official grounding, and clarifying UX—not bigger prompts alone.

Limitations

Competition-law domain only; not production legal advice without human review.