Workshop
Build an AI agent, connect it to patient data, and evaluate it against a rubric. The guided session takes 90–120 minutes and includes take-home artifacts.
Enter the workshopMedical AI benchmark toolkit
Build and evaluate medical AI
Join the hands-on workshop to build and evaluate an AI agent, or use the benchmark builder to turn a clinical question into a runnable, MAST-compatible benchmark.
Benchmarks
Choose a benchmark type, then work with a guided agent to define your clinical task, cases, references, and scoring for that benchmark.
Choose a benchmarkThree pieces of the stack, one workflow
Three pieces fit together. The model provides reasoning, the FHIR MCP layer gives the agent eyes and hands on patient data, and ARISEKit evaluates the output.
01.Design
Author the agent
Start from a clinical or admin archetype and write the agent Skill that defines how your agent behaves, all without leaving the browser.
02.Tools
Connect to FHIR data
Wire MCP-style tools so your agent can read and act on patient and operational data.The catalog is pre-built for workshop speed; the patterns transfer to production.
03.Evaluation
Score before you ship
Turn rubrics, test cases, and a judge model into a quantitative eval. Know whether your agent is actually good, not just impressive in a demo.