Front-line data, full-stack coding, the projects themselves.
An AI assistant that understands only one thing: DAS. I led the core bridge-monitoring algorithms and the product design. The decisive call came from the field — nobody pays for a standalone assistant; it has to live inside the monitoring screen customers already open every day. That call shaped the product.
Conversational AI for the health monitoring of major infrastructure. Two days to a working pipeline, two weeks to something I could demo to a customer’s CEO — brand, interface, manual and demo video all produced by me working alongside AI. The first vertical was a railway bridge, because that is where the engineering records are least forgiving.
There is no quiet place in a city, and that turns out to be good news. With beamforming plus ConvLSTM we localize and track several moving sources at once — every car on the street becomes a free seismic source. Noise is not the thing you filter out; it is the data.
It started with a phone number from an intermediary — no assistant, no introduction. I made the call, flew out myself, presented to twenty-odd specialists and convinced them to let us instrument their towers. Once the first data came back, one experiment had turned into a new product line.
Leading industrial R&D. Bridges, tunnels, dams, airports, wind towers, slopes — in all six I walked the same chain end to end: talk to the customer, win the pilot permit, design the experiment, deploy the sensors by hand, keep the data honest, turn the findings into something sellable. What that work taught me was not an algorithm. It was that the field never matches the drawing, and any plan whose author won’t go there gets overturned there.
I wrote a constitution for how I work with AI: one window manages, one window executes, acceptance is done by a fresh context, and I sit in the middle as the message bus. I later distilled the practice into murDrift — an engineering discipline that keeps long projects from quietly drifting off course. The site you are reading is what came out of it.
Full record of papers, patents and press → miao-yu.com