Agentic RAG for Legal Discovery: Autonomous Deposition Analysis

ABSTRACT

The legal discovery process involves manually cross-referencing millions of pages of depositions and emails to find logical contradictions. Standard search algorithms fail to connect multi-document narratives. We introduce 'Legal-Swarm,' an Agentic RAG architecture tailored for the judicial domain. A primary 'Investigator Agent' formulates complex legal queries, while subordinate 'Reader Agents' comb through vector databases to extract narratives. An adversarial 'Cross-Examination Agent' is then deployed to actively find contradictions in the retrieved timelines. In a controlled study of corporate litigation data, Legal-Swarm identified 94% of human-verified contradictions in a fraction of the time.