Introducing sens8ai

Physical AI is having its breakout moment: robots grasp, cars self-drive, world models rebuild space. It has learned the world. It hasn’t learned us.

Sensors are everywhere, and more arrive every year. Nearly half of US adults already wear one,¹ recording their movement, sleep, breathing and heart rhythm every second of every day. That creates two very different data problems: wearables produce a flood of continuous data, far more than any team can label, while each new sensor starts with almost none. Either way, the intelligence hasn’t caught up. That’s the layer we’re building.

The same rigid pipeline, every time

I’ve spent my career building AI on unstructured health data, and the way models get built hasn’t changed: collect the data, label it for months, train a model for one question, validate it, ship it. Every step is shaped around that one question, so the data and the model inherit its bias, and miss whatever else the signal could tell us. Then a new endpoint, a new question or a new kind of input comes along, and it’s back to the start of the same rigid pipeline, just to find out whether the idea works.

That pipeline was built for one sensor and one question at a time. It can’t keep up with either side: the flood from wearables, or the trickle from a new sensor. The need for models isn’t going away. The way we build them needs an overhaul.

The shift: Darwin™, one engine behind every answer

What it makes possible. Darwin puts one engine behind every question, so a new idea becomes an experiment, not a project. A sensor maker can test a new capability on the data it already has, before committing to build it. A research team can ask a new question, or add a new kind of signal, without rebuilding the pipeline first. That’s how teams prototype faster, uncover more of what each signal is saying, and turn life signals into early warnings, not surface-level metrics.

One unit for every life signal

Language AI was transformed when all text could be broken into one common unit, the token, so a single model could learn from all of it at once. Life signals never had that. The same heartbeat, from a chest strap, a ring or a hospital monitor, arrives at a different rate, in a different unit and format, so every model has had to start over with every sensor.

The Darwin Signal Unit™ (DSU) is the missing standard: one common unit for every life signal, whatever recorded it. With every signal in the same unit, a model can finally learn across sensors and questions together. That’s the shift language AI already had, and Darwin brings it to signal AI.

Language AI got the token.
Signal AI gets the DSU.