TARSE: Test-Time Adaptation via Retrieval of Skills and Experience for Reasoning Agents

Researchers developed TARSE, a new AI system for clinical decision-making that retrieves relevant medical skills and experiences from curated libraries to improve reasoning accuracy. The system performs test-time adaptation to align language models with clinically valid logic, showing improvements over existing medical AI baselines in question-answering benchmarks.

• →TARSE addresses clinical AI failures by retrieving procedural knowledge and prior examples rather than just facts.
• →The system uses separate libraries for clinical skills (guidelines, protocols) and experience (verified reasoning chains).
• →A step-aware retriever selects the most relevant skills and experiences for each specific medical case.
• →Test-time adaptation prevents AI reasoning from drifting toward unsupported medical shortcuts.
• →Experiments demonstrate consistent improvements over strong medical RAG baselines and prompting methods.