Skip to content
Let’s talk

ARISE, MAST, and medical AI research

Medical AI engineering benefits from careful evaluation of what models can do and where they fail. This page connects our founder’s public research profile with resources for understanding clinical AI evaluation.

Chase Walton and ARISE

ARISE’s public team directory lists Chase Walton, founder of C&C WaltonTech, among its members of technical staff, with Harvard University shown as his affiliation. ARISE brings together researchers, clinicians, and engineers working on healthcare AI.

The directory also lists investigators at Stanford University and Beth Israel Deaconess Medical Center (BIDMC) / Harvard University. These are research-network connections; C&C WaltonTech’s commercial services are offered independently.

What is MAST?

MAST stands for the Medical AI Superintelligence Test. ARISE describes it as a benchmark platform for evaluating medical AI across clinical capabilities. Its purpose is to examine performance across different tasks rather than reduce clinical competence to a single kind of question.

Read ARISE’s introduction to MAST and explore the MAST benchmark platform for the current methods and results. Model scores can change as evaluations and models change, so the original platform is the source for current comparisons.

Harvard, Beth Israel, and Stanford clinical AI research

One example of the network’s work is PACT, Physician-AI Collaboration Teaming. ARISE’s May 2026 announcement describes a Stanford-led project with collaborators at BIDMC / Harvard Medical School and Erasmus Medical Center. The project studies how AI supports clinical work and physician decision-making.

See the ARISE PACT announcement for the project scope and named investigators.

How to use benchmarks when building clinical software

A benchmark can help identify candidate models and questions to investigate. Turning that evidence into a software decision requires a precise description of the intended task.

  • Match the task. Identify whether the evaluation tests the same inputs and outputs as the proposed application. A text-only result cannot establish performance on an imaging task.
  • Inspect failure patterns. Review incomplete answers, unsupported statements, and cases that need human intervention. An average score may hide a recurring problem.
  • Keep comparisons reproducible. Record model versions, prompts, tools, evaluation criteria, and the conditions under which each result was obtained.
  • Evaluate the complete workflow. Test the data connection, interface, and review process together. Model performance alone does not establish how an application will work for its users.

Our medical AI integration and evaluation services focus on these engineering questions. To discuss a clinical workflow or evaluation problem, contact C&C WaltonTech.

Public research sources reviewed September 16, 2026. Institutional names describe the cited research network and projects.