Real-Time Intelligence for the Physical World

Every physical system speaks in energy. Energy Quotient is building the intelligence layer that comprehends it, starting with the electrical infrastructure under the most pressure today.

The easy descriptions of us are “AI for power quality” and “reinventing the oscilloscope.” Both are accurate, and both are too small. We are building the missing instrumentation and comprehension layer for the real-time physical world: a complete, continuous record of the electrical waveform, and physics-grounded AI that reasons over it with the physics in charge. The same architecture extends to every sensor and control signal that changes over time.

Energy Is the Common Language

Mechanical, thermal, chemical, and electrical systems all exchange energy, and in a modern facility nearly all of it passes through the electrical system. A bearing that is wearing, a heater that is short-cycling, a power supply that is failing, and a grid that is sagging each leave a signature on the voltage and current waveform. Read completely and continuously, the waveform is a running account of what every connected machine is doing.

Energy Quotient keeps that account whole: every sample, continuously, for as long as the sensor is installed. When a question comes up, the event that matters is already in the record, with everything that led to it and followed it. Today the same question usually means a technician arriving with a case of portable instruments, or a power quality logger left in place for a few weeks. That logger is still a triggered meter: it reports averages and keeps short clips when a threshold trips, so it can miss the event that matters, and no later software or inference can restore the context it never recorded.

Physics in Charge, AI in Service

General-purpose AI learns physics secondhand: from what people have written about it, sharpened by reinforcement learning on graded answers, never on live measurements of a real system. That works for describing a system and fails for operating one: a model never held to conservation of energy can be confidently wrong the moment conditions leave its training data, and an engineer cannot see why.

Physics-grounded AI reverses the arrangement. Deterministic signal processing and power-system analysis establish what the measurements show. AI interprets that evidence, explains it in plain language, and writes further analysis when a question calls for it. Engineers decide what to act on. Every conclusion traces back to waveform samples an engineer can check, which is what makes it trustworthy in infrastructure where a wrong action is expensive.

The next layer generalizes that analysis into an equation-based model of the system, run beside the measurements. Each component is described once, in equations drawn from circuit laws and conservation of energy, regardless of its boundary conditions, so the same components assemble into any topology and a finding on one system carries directly to another. Where the model and the measurements disagree, the difference is evidence: it locates the fault, the drift, or the change in operation. AI works on that difference, and the physics stays the arbiter. This architecture is the subject of U.S. Patent Application No. 19/556,943.

The Signal Layer Nobody Owns

Cameras have a mature AI stack, from the sensor to models trained on what it sees. The other signals machines measure and control over time have no equivalent: voltage and current, vibration, acoustics, pressure, flow, temperature, and the setpoints and commands of every control loop. Most of that data reaches AI as historian tags sampled every second or minute, already stripped of the dynamics that explain a failure, and the time-series models applied to it learn statistics, not physics.

Energy Quotient is building that missing layer: real-time AI that works at the signal level, on every sample at the sensor’s native rate, with the physics in charge. The electrical waveform comes first because nearly every machine’s behavior passes through it, and Energy Quotient already records it completely, at 32,000 samples a second on every channel. The continuous record, the residual against an equation-based model, and the reasoning over both carry to any of the others, given a model of its physics.

What We Are Building

EQ Syntropy runs on EQ-hosted servers or on a customer’s own edge AI hardware, and the language model on top is the customer’s choice of a frontier model or a self-hosted open-weight one. The analysis runs beside the data, so raw waveforms never leave the system, and with a self-hosted model the whole system can run air-gapped.

Why Now

Four forces are arriving at once, and each shows up first in the waveform.

Outages and power quality problems cost U.S. businesses an estimated $119 billion to $188 billion a year in 2001 (EPRI), and most power quality problems start inside the facility, where the operator can fix them once they can see them. Running existing capacity harder and more safely is a measurement problem before it is a generation problem. New generation takes years and billions of dollars to deliver its first watt. Visibility pays from the first installed sensor, and on a new build it pays across engineering, procurement, construction, and commissioning, before that first watt.

Where We Start, and Where It Leads

We start with reliability in facilities where a disturbance is expensive: semiconductor fabs, data centers, hospitals, defense installations, and advanced manufacturing. For them the facility is the grid’s last mile, where the most sensitive load lives and conventional instruments stop seeing.

The same record serves far more than one question. The capture that diagnoses a voltage sag at an imaging suite also shows a rooftop inverter’s harmonics, a motor beginning to fail, a demand peak worth shaving, and a battery string drifting out of balance. One measurement infrastructure and one reasoning architecture extend to every system that converts energy, and to the other sensors and control loops that watch it.

Every installation also adds to something no simulation supplies: years of gapless, real-world records of how electrical equipment actually behaves. That is the ground truth a physical model of the world has to be trained and validated against.

Built from Physics and Systems Software

Energy Quotient’s founders come to the problem from its two sides. Kevin Davies is a PhD engineer who models physical systems from first principles, from the Modelica Standard Library to production hybrid-vehicle control at Ford. Anthony Reffle is a software, systems, and AI engineer who architected the real-time Rust platform and the AI stack that reasons over its record. Between them, the measurement, the software, and the reasoning are designed as one system. Meet the team.

Shipping, Advancing, and Ahead

Here is what ships today, what we are building next, and where it leads.

You can operate the production platform yourself, on live data, in The Control Room.

The Control Room
Pilot Deployment
EQ Lab Kit

Start a Conversation

For partners, investors, and teams with a hard power problem: talk with the founders.