TheraAgent: AI Framework for Iterative Treatment Planning
TheraAgent introduces an innovative agentic framework designed for large language models (LLMs), shifting from one-shot generation to a systematic generate-judge-refine approach for treatment planning. In contrast to current LLMs that may create imprecise and unsafe plans, TheraAgent emulates the reasoning of human experts by iteratively enhancing drafts into accurate, detailed, and safer therapeutic strategies. Central to this framework is TheraJudge, a specialized evaluation module that is embedded within the inference loop to uphold clinical standards. This framework has set new benchmarks on HealthBench, a standard for healthcare AI. The findings are available on arXiv with the identifier 2605.05963.
Key facts
- TheraAgent is an agentic framework for LLMs.
- It uses an iterative generate-judge-refine pipeline.
- TheraJudge is a treatment-specific evaluation module.
- The framework achieved state-of-the-art results on HealthBench.
- The research is published on arXiv (2605.05963).
- It addresses limitations of one-shot generation in LLMs.
- The approach mirrors human expert reasoning.
- It aims to produce safer treatment plans.
Entities
Institutions
- arXiv