Keywords: RAG, Agentic, Biomedical, clinical, retrieval, pubmed, mimic
TL;DR: This paper introduces an agentic RAG framework with dual domain-specific text encoders that iteratively refines queries to achieve competitive performance on the PMC-Patients and PubMedQA benchmarks for biomedical evidence retrieval.
Abstract: We propose an agentic RAG framework for biomedical evidence retrieval that uses iterative query refinement across PubMed and MIMIC-IV clinical notes. Using dual domain-specific encoders and self-critique loops, our system achieves competitive results on PMC-Patients and PubMedQA benchmarks, demonstrating the value of adaptive retrieval for clinical decision support.
Submission Number: 17
Loading