A retrieval-augmented multi-agent system creates evidence-based, fine-grained rubrics for medical LLM evaluation, achieving 50.20% and 31.90% CIA scores on HealthBench and LLMEval-Med while outperforming GPT-4o baselines.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Automated Rubrics for Reliable Evaluation of Medical Dialogue Systems
A retrieval-augmented multi-agent system creates evidence-based, fine-grained rubrics for medical LLM evaluation, achieving 50.20% and 31.90% CIA scores on HealthBench and LLMEval-Med while outperforming GPT-4o baselines.