LexisNexis prioritizes AI completeness in legal applications, using metrics like comprehensiveness, authority, and hallucination rates beyond mere accuracy. They've advanced from standard RAG in Lexis+ AI to graph RAG and agentic graphs in Protégé, incorporating planner and reflection agents. This approach manages uncertainty for reliable high-stakes outputs.
Key Points
- 1.Establishes sub-metrics for usefulness: authority, citation accuracy, hallucination rates, comprehensiveness
- 2.Lexis+ AI (2023) uses standard RAG with hybrid vector search
- 3.Protégé (2024) adds knowledge graph layer atop vector search for authoritative answers
- 4.Planner and reflection agents parse requests and self-critique outputs
Impact Analysis
Enhances trust in legal AI by addressing partial answers and non-citable sources, influencing enterprise standards for high-stakes gen AI evaluation and deployment.
Technical Details
Evolves to graph RAG and agentic graphs beyond standard RAG; knowledge graph overcomes semantic search limits on authority. Planner agents parse queries; reflection agents critique outputs for quality.

