Nabz · A Roshan AI model family

Clinical reasoning on the signals themselves.

An intensive care unit produces some of the densest data in medicine, and almost all of it is consumed as a number on a monitor or a black-box alarm. Nabz is our family of models for that stream. Each one reads the waveform directly and returns something a clinician can interrogate down to the window that caused it.

4
Signal modalities
ECG · PPG · ABP · RESP
2
Models
More in research
100%
Traceable to signal
By construction
What makes it a family

Three rules every Nabz model follows.

The signal is the source

Every Nabz model reads continuous physiological waveforms directly. Nothing in the family depends on a summary statistic somebody else computed first.

Structure is earned, not declared

What a model asserts about a patient has to be supported by a measurable property of that patient’s signal. An ontology author does not get to decide it in advance.

Traceable to the waveform

Any claim a Nabz model makes can be opened down to the millisecond-level windows behind it. The audit path is part of the architecture.

Nabz sits alongside existing monitoring and the EMR. These models detect physiological concepts and their relationships, not disease labels, and the deployments described here are retrospective and research-oriented.

Nabz · open to collaboration

Bring your signals.

We are extending the concept vocabulary, adding modalities, and looking for retrospective cohorts to validate against. Research collaboration is welcome on every model in the family.

Start a conversation