The signal layer of physical AI

Sensors are everywhere.
Signal isn't.

We turn raw real-world sensor streams into measurements, detections, and predictions that hold up in the clinic, on the line, and in the field, across physiological, optical, industrial, volumetric, and time-series data.

The signal is already there. It is just too faint to read. We recover it.


What we do

What we do, and two ways to work with us

For tier-1 medical centers and deep-tech companies where the sensor data is the hard part. The signal is faint, the data high-dimensional, and the domain unforgiving.

The Signal Layer

Deep-tech signal recovery that turns messy optical, physiological, volumetric, industrial, and time-series sensor streams into reliable detections and predictions that hold up in production.

Physiological · optical · industrial

Research Partnerships

We co-develop with tier-1 medical centers: measurements and markers from raw physiological and imaging data. Co-publish, build the tools.

Co-develop · co-publish

Technical Leadership

We bring our signal-recovery stack and take hands-on technical leadership inside deep-tech teams where the raw sensor data is the hard part.

Startups · industrial AI · defense tech
A noisy sensor trace on the left becoming a clean signal with three marked readings on the right.

From raw stream to readout.

Our work so far

Paid research partnerships with tier-1 medical centers.

Measurements and markers from raw physiological and imaging data, validated on real patient data.

Production AI for companies in optical and physiological sensing.

Built hands-on inside their teams, and holding up in production.

Peer-reviewed, with our partners as co-authors.

A paper at OMIA 2026, the ophthalmic imaging workshop at MICCAI.

Publications

How we work

Production-grade AI for the physics of weak signals.

Signal recovery, production AI, and clinical and field validation, in one lab. Founded by Michael Chertok, 20 years of production AI, ex-Amazon Principal Scientist.

Depth, not volume.

We take on a few problems at a time, where the signal is faint and the domain unforgiving, and we stay until the model holds up in production.

The signal is already there.

We start from the sensor's raw output, before the device's preprocessing has thrown it away.

Hands-on, inside the team.

We bring our signal-recovery stack directly to the principal investigator or the engineering team that owns the sensor, validating against real ground truth from first readout to deployment.

Write to us

We take on problems where three things meet: an untamed physical signal, a domain expert who knows what it means, and a decision that waits on the readout. If that is what you are working on, write to us.