From doctor to the building side of generative AI.

Born into a family of doctors

My family runs a clinic in Wakayama. I was born into a line of doctors — grandfather, father, uncle, and my younger brother and sister too. In high school I was stronger in the humanities, and my homeroom teacher tried to stop me: "With those grades, going into the sciences won't guarantee you make it as a doctor. It's a thorny road." That one line lit a fire in me instead. After two years as a ronin in Osaka, I got into the medical school at Oita University.

I liked Oita, so after graduating I stayed and became a resident at a hospital in Beppu. I passed the 116th National Medical Licensing Exam in 2022. I assumed I'd simply go on being a doctor.

The summer it hit me like a punch

Summer 2022, my first year as a resident. August 23, September 2 — I still remember the dates. Stable Diffusion dropped. The moment I saw it, it hit me like a punch to the head: "This is a different paradigm, a revolutionary technology."

The trigger wasn't medicine — it was my camera hobby. I'd been shooting landscapes and people, and I thought, "I don't even have to go on a shoot; I can just make this myself." But the models back then were skewed toward Western data and couldn't render Asian faces well. Desperate to generate good-looking Japanese faces, I started combing overseas forums for fine-tuning techniques. It was pure otaku obsession, an extension of a hobby. That November ChatGPT arrived, and the hunch hardened into conviction: "This is the biggest revolution since the internet."

In Oita I also worked as a photographer; the work is still on ikora-128.com. Diaries and the record of leaving medicine went on longisland3.com. I don't update either now, but that version of me is still there.

The model went viral, and I got headhunted

In January 2023 I finished a model that could generate Asian women cleanly. When I posted it on forums and social media, the scene blew up — people started trying to figure out who I was. Companies and VCs flooded me with offers to "do research together" or "start a company," but I turned them all down: "My real job is being a doctor."

The turning point was an intense pitch from the CEO of Airfriend — a chat app for talking and calling with AI characters, on the order of a million users. He was the author of a deep-learning blog I'd read and admired back in med school, a mentor in my head. "Being a doctor is a waste. You should quit," he told me flatly. With a "if it doesn't work out, I can go back to medicine in a year" mindset, in my second year of residency I took leave from the hospital and joined as an image-generation AI engineer.

From unemployed to CTO

I'd jumped in, but at the time I didn't even know how to SSH. The company was fully remote, and I kept building alone from my home in Beppu, never once meeting a teammate in person. The only thing I could lean on was ChatGPT — I asked it everything and caught up through technical books and overseas YouTube.

About a year in, I left over a difference in direction and — with nothing lined up — briefly became "unemployed." Using an engineer café in Fukuoka as a base, I decided to show up to every drinking party I was invited to. That's how I connected with Dr. Satoshi Kodera, who leads medicine × generative AI in the cardiology department at UTokyo Hospital. "Come join us. The one condition is that you move to Tokyo," he said. "I'm unemployed right now, so I'll be there next month," I answered on the spot.

I moved to Tokyo in June 2024 and held two roles at once — research staff at UTokyo Hospital and Director & CTO of the AI startup Livetoon. At Livetoon I built the speech synthesis (TTS) from the architecture up, entirely from scratch. At the time only about five companies in Japan — big names included — could do this in-house. As the lead of a four-person team, I finished "Livetoon TTS," among the best Japanese TTS around. Image, speech, LLMs — I think I've covered nearly every area of generative AI except video and robotics.

If no one else will, I will

I'd deliberately steered clear of medicine, assuming it would turn into a red ocean, and built AI in entertainment instead. Surely the brilliant people would quickly make AI that rescues the clinical floor. But three years passed and no player showed up to solve the fundamental problems.

The deciding factor was the cries for help from my peers and my sister, all practicing doctors. "Charts and paperwork are genuinely inefficient. Can't AI do something about this?" "The DX vendors who come in from outside don't get the feel of the floor, and it turns into a fight. You do it." In January 2026 I founded GENSHI AI and became its CEO. Auto-drafting discharge summaries, sovereign-AI speech recognition that keeps data inside the hospital — I'm building from the problems on the floor, in order. That same year, we were selected for AMED's support program.

I'm not going back to medicine now. "Can't go back" might be more accurate. I'm standing at a decisive moment for how Japanese medicine changes, and honestly I feel this is no time to be practicing medicine. If someone better than me shows up, I'll hand it over anytime. But for now, no one else is doing it.