Vancouver · Canada · 49.2° N · Applied AI for the physical world
Coriolis Labs builds AI that gathers a world of imperfect signals into one living, calibrated picture, and turns it into decisions fast enough to matter and honest enough to replay.
The thesis
We build the intelligence in between: systems that watch many imperfect feeds at once, reconcile them against the physics of the real world, hold a living estimate of what is happening, and act on it, with calibrated confidence, honest uncertainty, and a complete evidentiary trail. Not a dashboard on top of the noise. A mind for the environment itself.
The system · one path, four commitments
Cameras, radios, instruments, records: our systems take heterogeneous signals as they arrive and reconcile them against the physics of the real world, so conflicting or degraded sources sharpen the picture instead of breaking it.
A carried belief holds through dropout, occlusion and noise. Uncertainty grows honestly when a signal disappears and settles the moment it returns; a quiet feed is never mistaken for a quiet world.
Calibrated calls in milliseconds, with confidence you can read and the discipline to abstain when evidence is thin. People set the policy; the system honours it faster than any queue of humans could.
Full provenance from raw signal to output. Any alert, forecast or action can be reconstructed bit for bit and audited after the fact, by an operator, a regulator, or a court.
How it learns
The heavy learning happens offline: from complete histories, physical models and simulated consequence, in an environment we govern end to end. What ships is small: a distilled system that runs at the edge, on hardware already in place.
No cloud dependency. No assumption of connectivity. It keeps working when the network doesn't, which, in the environments we care about, is precisely when it matters most.
Where it matters
Dual-use from day one. The same core that helps a city see itself serves the missions where the stakes are highest.
The feeds a city already publishes become early warning and foresight, without watching individuals.
Water, power, transit, telecom: drift and anomalies caught while they are still small and cheap.
Weather, water and air fused into operational foresight for the seasons that test a region.
Plants, ports and fleets that notice their own degradation before it becomes downtime.
Sparse signals over vast distances: understanding that persists between contacts.
Decision-grade awareness under contest, built and controlled to sovereign standards.
How we work
Preregistered tests, held-out truth, independent review. A claim stays blocked until the data exists to support it, including our own.
Our systems say "not enough evidence" out loud. Absence of signal is never read as absence of risk.
We measure environments, not people. Minimization is structural, not a setting; identification of individuals is excluded by design.
Built and controlled in Canada, from the training pipeline to the deployed edge, with nothing critical surrendered to a supplier.
Humans set the terms and keep the final word. Machines honour those terms at machine speed.
Start the conversation
We are at the beginning, building openly with partners across cities, industry and government. If your world produces more signal than judgment, we should talk.
briefings@coriolislabs.ca