MARC v1: Open-Source Multi-Agent Framework for Clinical AI Reasoning

MARC v1 has been introduced by researchers as an open-source framework aimed at improving clinical AI reasoning through the orchestration of multiple agents. This framework, outlined in a paper available on arXiv, replaces traditional monolithic LLM prompting with a deterministic system featuring role-specific agents responsible for extraction, reasoning, answer generation, and evaluation. This method facilitates explicit context transfer and allows for traceable intermediate outputs, which aids in identifying failure points. Central to this is the Decomposer module, which creates task-specific agent prompts from simple language descriptions, removing the necessity for manual prompt engineering. MARC is deployable both via API and on local CPUs, fully configurable through YAML without needing code alterations. It is model-agnostic and interpretable, making it accessible to clinical experts without programming skills. The complete framework can be found online. The paper was submitted to the Artificial Intelligence section of arXiv and includes details on submission history and citation tools. This framework reflects a growing movement towards modular and interpretable AI systems in healthcare, focusing on enhancing reliability and transparency in clinical decision-making.