SEMA-RAG: A Self-Evolving Multi-Agent Retrieval-Augmented Generation Framework for Medical Reasoning

2025-12-28 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, long

Summary

SEMA-RAG, a Self-Evolving Multi-Agent Retrieval-Augmented Generation framework, is proposed for medical question answering to address limitations of static, single-round RAG in clinical reasoning. Standard RAG often suffers from poor question-to-query translation and a lack of iterative sufficiency feedback, leading to unreliable evidence chains. SEMA-RAG decouples these tasks into three specialized agents: an Interpreter Agent for clinical schema interpretation, an Explorer Agent for sufficiency-driven self-evolving retrieval, and an Arbiter Agent for evidence adjudication and answer selection. Evaluated across five medical benchmarks (MMLU-Med, MedQA-US, MedMCQA, PubMedQA*, BioASQ-Y/N) and five LLM backbones (deepseek-v3.1, kimi-k2, qwen3-coder-plus, gemini-2.0-flash, glm-4.0-flash), SEMA-RAG consistently improved the strongest baseline by an average of +6.46 accuracy points per backbone. The framework's multi-agent architecture and iterative evidence exploration are crucial for its performance gains.

Key takeaway

For AI Scientists and Machine Learning Engineers developing RAG systems for high-stakes domains like healthcare, consider adopting a multi-agent, self-evolving framework. Your current single-round RAG approaches may be insufficient for complex reasoning tasks, leading to suboptimal accuracy. Implementing task decoupling and iterative, sufficiency-driven retrieval can significantly enhance evidence chain reliability and overall system performance, as demonstrated by SEMA-RAG's +6.46 accuracy point improvement.

Key insights

Multi-agent RAG with self-evolving, sufficiency-driven retrieval significantly improves medical question answering accuracy.

Principles

• Decouple complex reasoning tasks into specialized roles.
• Iterative retrieval with sufficiency feedback enhances evidence chains.
• Clinically grounded semantic interpretation improves query quality.

Method

SEMA-RAG uses an Interpreter Agent to structure questions into a clinical schema, an Explorer Agent for self-evolving, multi-round retrieval based on evidence sufficiency, and an Arbiter Agent to adjudicate evidence and select answers.

In practice

• Implement role-based agents for complex RAG tasks.
• Use iterative query refinement to address evidence gaps.
• Structure input questions into explicit schemas for better retrieval.

Topics

• SEMA-RAG
• Multi-Agent RAG
• Medical Reasoning
• Retrieval-Augmented Generation
• Clinical Schema Interpretation

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• From Conflict to Consensus: Boosting Medical Reasoning via Multi-Round Agentic RAG — MA-RAG (Multi-Round Agentic RAG) is introduced to enhance medical reasoning in Large Language Models (LLMs) by mitigating issues like hallucinations and outdated knowledge, which traditional Retrieval…
• Beyond the Blank Stare: Injecting Medical Expertise into LLMs Through RAG-Powered Knowledge — RAG systems enhance LLM reliability in medicine by grounding responses in verifiable, external knowledge bases.
• LLM-Based Multi-Agent System with Retrieval-Augmented Generation for Medical Care Planning Generation in Sickle Cell Disease — RAG-enhanced multi-agent LLMs can improve safety and clinical quality in high-risk medical care planning.
• Agent-Orchestrated Adaptive RAG: A Comparative Study on Structured and Multi-Hop Retrieval — Agentic RAG enhancements must be selectively applied based on query and domain characteristics for optimal performance.
• Why Retrieval-Augmented Generation Fails: A Graph Perspective — RAG failures stem from shallow, fragmented internal information flow, not just poor retrieval.
• TechGraphRAG: An Agentic Graph-Augmented RAG Framework for Technical Literature Reasoning — Agentic RAG frameworks can enhance technical reasoning by integrating multi-step pipelines, knowledge graphs, and external search for robust evidence grounding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.