Tokyo, Japan · San Francisco Bay Area
Yu Akagi, MD
Physician · Researcher · AI engineer
PhD student at the University of Tokyo
Vision 01
Automate clinical workflow with AI, for both clinicians and patients.
Physicians spend most of their working hours in the electronic health record. In other words, they are at a computer for much of the day. AI has now advanced far enough to operate a computer the way a person does. My research aims to use AI to make that computer work more efficient, so that physicians can give more of their attention to the patients in front of them.
Vision 02
Build general-purpose generative models from large-scale real-world clinical data.
Hospitals record every admission, prescription, and test. In other words, they hold a detailed account of what actually happened to millions of patients. Most of it cannot be used yet, because every hospital stores its records differently. My research aims to standardise those records and train generative models on them, so that we can simulate how a patient's illness is likely to unfold.
Selected work
Research systems for simulating and automating clinical work.
Summer 2026Concept & lead engineerA world of AI agents
Agentic systems usually hide behind a chat box. We built a game-like world you can walk into instead, as a case study in AI for materials science at Lawrence Livermore National Laboratory.
PythonJavaScriptNext.jsTailwind CSS
First author & lead engineerWatcher
A simulator for patient trajectories. Given a patient's medical history, it generates possible futures. Trained on more than 200 million real clinical records.
PythonPyTorch
Blog
Notes on research, engineering, and whatever else is occupying me.
Why I do AI for healthcare
What I learned from my patients.
We built a town for our AI agents
What happens when you take an agentic system out of the chat box and put it in a game-like town instead. Built over a summer internship, in two minutes of video.
Living in Livermore
What it is like to live in Livermore, California: shopping, getting around, and where to go nearby, for whoever comes next.
