MARC v1: An Open-Source Multi-Agent Framework for C...

MARC v1: An Open-Source Multi-Agent Framework for C... | AI Research

Key Takeaways

MARC v1: An Open-Source Multi-Agent Framework for Clinical AI Reasoning and Coordination introduces a framework designed to improve clinical AI by replacing...
We present Multi-Agent Reasoning and Coordination (MARC), an open-source framework that replaces monolithic LLM prompting with deterministic multi-agent orchestration for clinical reasoning.
MARC coordinates role-specialized agents for extraction, reasoning, answer generation, and evaluation, with explicit context passing and traceable intermediate outputs, enabling stage-wise failure attribution.
We additionally introduce a Decomposer module that generates task-specific agent prompts from a plain-language description, eliminating manual prompt engineering.
The framework supports both API-based and local CPU-compatible deployments and is entirely configurable via YAML, without code modifications.

Paper AbstractExpand

We present Multi-Agent Reasoning and Coordination (MARC), an open-source framework that replaces monolithic LLM prompting with deterministic multi-agent orchestration for clinical reasoning. MARC coordinates role-specialized agents for extraction, reasoning, answer generation, and evaluation, with explicit context passing and traceable intermediate outputs, enabling stage-wise failure attribution. We additionally introduce a Decomposer module that generates task-specific agent prompts from a plain-language description, eliminating manual prompt engineering. The framework supports both API-based and local CPU-compatible deployments and is entirely configurable via YAML, without code modifications. MARC is designed to be model-agnostic, interpretable, and accessible to clinical domain experts without programming expertise. The full framework is available at this https URL .

MARC v1: An Open-Source Multi-Agent Framework for Clinical AI Reasoning and Coordination introduces a framework designed to improve clinical AI by replacing monolithic large language model (LLM) prompting with a deterministic, multi-agent system. By breaking down complex clinical tasks into specialized, traceable steps, the authors aim to make AI reasoning in healthcare more interpretable and reliable.

Orchestrating Clinical Reasoning

The MARC framework coordinates a team of role-specialized agents to handle distinct stages of a clinical task: information extraction, reasoning, answer generation, and evaluation. Unlike traditional methods that rely on a single, large prompt, MARC uses explicit context passing between these agents. This structure allows for "stage-wise failure attribution," meaning that if an error occurs, users can trace exactly which agent or step in the process failed.

Automating Prompt Engineering

A core component of the framework is the "Decomposer" module. This tool allows users to provide a plain-language description of a task, which the module then converts into task-specific prompts for the agents. This design is intended to eliminate the need for manual prompt engineering, making the system accessible to clinical domain experts who may not have programming experience.

Deployment and Flexibility

MARC is designed to be model-agnostic, meaning it can function with various underlying LLMs. The framework is entirely configurable through YAML files, allowing users to modify the system without changing the underlying code. It supports both API-based deployments and local, CPU-compatible environments, providing flexibility for different technical infrastructures.

Franklin Analysis

The design of MARC suggests a shift toward modularity in clinical AI. By prioritizing interpretability and removing the need for manual coding, the framework addresses common barriers to adopting AI in medical settings. The ability to perform stage-wise failure attribution is a significant feature for clinical environments, where understanding the "why" behind an AI's output is critical for safety and verification. Because the framework is open-source and configurable via YAML, it is positioned to be a practical tool for researchers and clinicians looking to implement structured, multi-step reasoning without building custom architectures from scratch.