RareDx: Building a multi-agent AI system to transform rare disease diagnosis

This article presents RareDx, a multi-agent AI system built on Amazon Bedrock AgentCore that transforms rare disease diagnosis by automating administrative workflows for physician geneticists.

• Orchestrates 12 specialist agents optimized for different diagnostic tasks like phenotype analysis, variant interpretation, and report generation
• Processes scanned referral documents, extracts clinical history, maps symptoms to HPO codes, and identifies diagnostic gaps
• Integrates 30 biomedical tools via Model Context Protocol including ClinVar, gnomAD, PubMed, and clinical trial databases
• Provides six-stage diagnostic journey interface with ambient listening, facial analysis, pedigree construction, and automated report generation
• Runs entirely in EU region with zero operator access, supporting GDPR and HIPAA compliance
• Initial testing showed 60-70% reduction in administrative burden, freeing physician time for clinical reasoning

RareDx demonstrates how modular multi-agent architecture can accelerate rare disease diagnosis while keeping physicians focused on patient care and clinical expertise.