About — Miao YU

I was trained to listen to the Earth.

For my doctorate in Paris I worked on full-waveform inversion: pushing seismic waves back into the ground, again and again, until invisible rock layers took shape. Those years taught me exactly one skill that mattered — how to look at a mess of vibrations and decide which part is information and which part is only noise. It has followed me everywhere since.

I did not stay in pure academia after graduating. From 2022 I helped start a seismic-sensing company and spent the next four years as its founding head of R&D. The reason for the move was plain: the physics a few kilometres underground is the same physics as the bridges, tunnels and dams above it. Cities are full of structures ageing slowly, and they have been vibrating all along — nobody was listening carefully.

For most of those four years I was not at a desk. Talking to customers, applying for pilot permits, designing the experiments, climbing bridges and towers to place the sensors myself, keeping batch after batch of data in order, then turning the findings into something a customer would pay for. Once it was a road collapse; I got to the site and stood at the edge of the hole. On a résumé all of that is a single line, but it changed how I make technical decisions: the field never matches the drawing, and any plan whose author won’t go there gets overturned there.

In the autumn of 2025 I left the company, returned to a university as a visiting scholar, and started building my own projects. What really turned me was an experience in June 2026: I took apart, line by line, a set of open-source tools that impose rules on AI agents — and suddenly saw it. An agent’s reliability does not come from how clever the model is; it comes from the structure of constraints around it. Which is uncannily close to what I had been doing for a decade: using structure to keep the trustworthy part of an uncertain signal.

Since then almost every project of mine has been built with a team of AI windows: one manages, one executes, acceptance goes to a fresh context, and I sit in the middle as the message bus. I distilled that discipline into murDrift. The questions I care about changed shape too — no longer only “how do I get this waveform right,” but “how do one person and a group of AIs build something complex, reliably.”

So this site is not a résumé. The complete papers, patents and press live on the main site; that is the archive, the part that has stopped moving. What lives here is the part still in motion: how I think now, what I am building, and which of my judgements the field has revised.

If you are working on vibration, sensing, or on collaborating with AI, write to me.