← back
03SIIM Hackathon · 2nd Place
Dream Team
LLM agents × MCP × specialist access
Find your dream team of global medical experts for your health condition, through an agentic LLM workflow.
With Joey Hentel and Kurt Teichman · SIIM 2025
The problem
Over 100 million people in the U.S. face barriers to reaching medical specialists. 20% of Americans live in rural areas, but only 9% of physicians practice there — so a patient handed a report reading “left renal mass” may have no qualified expert to ask.
What we built
- An orchestrated system of autonomous LLM agents that covers the whole referral journey, from report to booked consult
- Imaging access through DICOM MCP — agents can query DICOM servers for studies, series and image instances, and analyze clinical metadata in context
- Tool access through the Model Context Protocol, one unified access point instead of separate authentication and custom integration logic for every system
How it works
- Diagnosis agent — extracts the key findings from a radiology report and classifies them into an actionable category, like kidney cancer
- Referral agent — queries an expert database and matches the patient to a specialist by disease category, location, and preferences like experience and rating
- Secretary agent — checks availability with scheduling tools and sets up a video consultation
- Research + summary agents — produce a shareable consultation plan, including the latest research on the key findings
Technologies
LLM agentsModel Context ProtocolDICOM MCPScheduling tools
Screenshots
screenshot
screenshot
Outcomes
- A repeatable end-to-end workflow: summarize → classify → query → schedule → synthesize
- Presented with a live demo
- 2nd place at the SIIM Hackathon