Autonomous Multi-Agent Pipeline for Biomedical Research Automation

Abstract: This paper presents a multi-agent system for biomedical question answering that emulates a structured research workflow using large language models (LLMs). The proposed architecture leverages a LangGraph-based pipeline in which multiple specialized agents collaboratively perform literature retrieval, hypothesis generation, experimental protocol design, and validation. Three parallel junior agents gather evidence from diverse biomedical sources, followed by a supervisor that synthesizes a unified hypothesis. Subsequent agents refine the hypothesis, design experimental protocols, and conduct peer and safety reviews through an iterative feedback loop. A principal investigator agent produces a final decision, while an evaluator module acts as an LLM-as-a-judge to assess quality, precision, recall, latency, and cost. The system integrates retrieval-augmented generation (RAG), structured JSON outputs, and retry mechanisms to ensure robustness and consistency. Experimental evaluation demonstrates that the proposed approach improves reasoning depth, factual grounding, and decision reliability compared to single-agent baselines. Additionally, the framework provides transparent cost and latency tracking, making it suitable for real-world research assistance applications.

Keywords: Multi-Agent Systems, Biomedical Question Answering, Large Language Models, Retrieval-Augmented Generation, LangGraph, Experimental Protocol Design, LLM Evaluation, AI in Healthcare.