Binya Liu · Physician-Scientist · Clinical AI

Making clinical AI safer, smarter, and more human.

Drawing on clinical medicine, research, and teaching, I develop rigorous AI evaluations and practical tools to support better care and healthier, happier lives.

Science for Health. Joy for Life.

Explore my work →

Selected Projects

Research meets real-world impact.

Clinical AI Benchmark

Prediabetes AI Benchmark

Can AI help people understand when to get screened, what to do next, and when to seek urgent care? I designed 100 synthetic patient scenarios to find out.

Explore the benchmark →

In Development

Prediabetes Screening App

A planned patient-friendly health application exploring screening guidance, risk communication, and safer next steps.

Coming soon
Concept only — not a functioning medical application.

Featured Research · Prediabetes AI Benchmark

Earlier Awareness Better Decisions Healthier Lives

Prediabetes often has no obvious symptoms. But early screening and healthier habits can help prevent or delay type 2 diabetes.

115.2M

U.S. adults estimated to have prediabetes

CDC · 2026

8 in 10

U.S. adults with prediabetes who don't know they have it

CDC · 2026

92.5%

AI agreement with predefined next steps in 80 held-out synthetic cases (74/80)

JoyBinya · Independent benchmark

The first two figures are U.S. public-health estimates from CDC (2026) ↗ . The third is a benchmark result, not clinical accuracy.

Why I Built It

As more people turn to AI for health information, I wanted to know whether a GPT-based model could recommend appropriate screening, judge urgency, recognize missing information, and explain the next step safely.

What I Found

Across 80 unseen cases, 74 recommendations matched the predefined next steps. The model also classified all 8 cases in a small urgency check. Physician review showed why safe clinical guidance cannot be reduced to one score.

My Vision

Help build trustworthy AI that encourages earlier awareness, appropriate screening, timely care, and informed choices for healthier lives.

This evaluation used synthetic scenarios. Its findings do not establish real-world clinical accuracy, safety, or improved patient outcomes. The benchmark is not a diagnostic tool or medical advice.

About Me

From Clinical Medicine to Healthcare AI

Medicine has shaped my life since I entered medical school at 17. Years of patient care, research, and teaching have taught me that great healthcare requires sound judgment, scientific rigor, and genuine empathy.

Today, I bring these perspectives to healthcare AI — building trustworthy evaluations and practical tools that support clinicians, empower patients, and help more people make informed decisions about their health.

~10 Years Clinical Training & Patient Care

10+ Years Medical Research

3+ Years Medical Teaching

Research & Publications