A convenience-store refrigerator is a modest machine with an immodest schedule. It hums through breakfast, lunch and the late-night dash for something cold, indifferent to the wholesale price of electricity or the strain on a transformer several blocks away. Matthew Irvin looked at that refrigerator and saw a clock. Cool the drinks a little earlier, before an expensive spike in demand, and the cold itself becomes a small reserve of energy. The cans stay cold. The store keeps operating. The peak arrives to find less work waiting for it.
That experiment, run by Irvin's Maplewell Energy at convenience-store sites in Denver and Salt Lake City, is the compact version of his larger bet. The electric grid needs new generation and new wires, but it also needs better choreography. Buildings, batteries, solar arrays and industrial equipment already contain thousands of decisions about when energy is used. Make those decisions predictive, automatic and measurable, and demand stops behaving like weather. It becomes something an operator can shape.
He learned the grid from the heavy end
Irvin did not arrive at climate software by way of a dashboard. He is a mechanical engineer and a licensed Professional Engineer in Colorado. His career included work at General Electric and Stanley Consultants, where the assignments were satisfyingly physical: design engineering, feasibility studies, plant condition assessments, controls, decommissioning plans, performance testing and tuning. In 2018, while based in Stanley's Denver office, he co-wrote an analysis of the choices facing Midwestern utilities as aging plants, transmission costs and changing generation economics squeezed their plans.
That work dealt in turbines, permits, capital schedules and the stern arithmetic of reliable power. A municipal utility could keep buying from the market, build ten megawatts now and ten later, or finance a larger plant. Every option carried a price, a delay and a different kind of exposure. The experience left a clear fingerprint on Irvin's later work: technology is only useful when it survives an owner's operating constraints and financial questions.
Maplewell, co-founded in 2019, turned his attention from deciding what generation to build toward deciding how existing demand should behave. Its core product, JANiiT, forecasts loads and prices, optimizes possible actions, and controls distributed assets in real time. Batteries are obvious participants. HVAC, refrigeration, solar and other flexible equipment can join the conversation too. Irvin co-created the optimization and predictive-control technology and has served as lead architect across deep-learning forecasting, cloud software and edge-computing architecture.
The useful loop: from tariff to verified response
Cold drinks, clean proof
The refrigerator pilot made the abstraction visible. JANiiT pre-cooled bottled and canned drinks before a predicted energy peak, using the thermal mass already sitting behind glass doors. It also coordinated HVAC. Maplewell reported a 25 percent reduction in peak energy use during the beta deployment, with savings flowing to the building operator. The company's separate projection of a 42 percent reduction assumed additional energy storage, a useful distinction between a field result and a modeled next step.
There is a product lesson hiding among the soda bottles. Maplewell began with one costly interval, found a load that could move without changing the customer's experience, closed the control loop and measured the result. The software did not ask a cashier to become an energy trader. It respected the store's actual job: keep products cold and customers served.
“Most building management systems lack any sort of dynamic demand management capability for controllable loads.”Matthew Irvin, on the refrigeration pilot
A power plant made of appointments
Coordinate enough local choices and the industry calls the result a virtual power plant. The name is vaporous; the ingredients are not. One customer has a battery. Another has rooftop solar. A third can shift cooling for half an hour. Software aggregates those small reserves and dispatches them when a utility or wholesale market needs relief. The collection can lower a customer's demand charge, earn grid-service revenue and help a utility avoid firing up expensive peaking supply.
Irvin has described two customers for this approach. Some organizations already own distributed energy resources and want more value from them. Others are considering storage but need several revenue streams to make the investment work. A battery can reserve capacity for resilience, shave a building's own peak and still participate in a wider grid program. The trick is allocating those promises without double-booking the electrons.
In 2023, Maplewell paired with Urban Electric Power, a maker of rechargeable zinc batteries, to pursue commercial and industrial virtual power plants. Irvin said the combination could help utilities alleviate distribution peaks and defer selected upgrades. His emphasis stayed behind the meter, close to the customer's economics and operations, even as the aggregated effect reached outward to the grid.
The cloud has to survive the real world
Energy control becomes more interesting when the network misbehaves. Maplewell's work for the U.S. Navy, underway since 2021, is designed for shipboard and shipyard environments where communications may be disrupted, disconnected, intermittent or low-bandwidth. JANiiT can run locally and air-gapped while still supporting remote monitoring through constrained links. This is where “edge computing” stops being a conference phrase and becomes an operating requirement. A control system cannot shrug because the cloud is having a difficult afternoon.
Irvin and Maplewell engineer Marcus Caudill were named principal investigators on the company's Phase II Navy work. The platform's remit included online optimization of energy storage and possible applications in hybrid-electric drives, directed energy and non-wired upgrades to ship bus capacity. The setting sharpened Maplewell's general product thesis: forecasting and optimization should travel to the asset, not depend on a perfect path back to a distant data center.
Another piece arrived from nearby NREL. In 2022, Maplewell secured an exclusive license to OptGrid, optimization technology developed through the Energy Department's ARPA-E NODES program. Maplewell said the technology could help JANiiT manage as many as one million energy devices. The partnership also connected Irvin with Emiliano Dall'Anese, the OptGrid inventor, who joined Maplewell as chief technology officer and later became a senior adviser.
The company accumulated a series of public checkpoints around the engineering. Maplewell won the energy and utilities category at the 2021 BizWest IQ Awards. Colorado's Office of Economic Development and International Trade awarded it a $250,000 Advanced Industries grant. In 2023, JANiiT received a Product Award in the Internet of Things category. None of those plaques could flatten a demand curve, but each marked a different kind of test: regional relevance, a case for commercialization and recognition from product practitioners.
Irvin also had to assemble people whose specialties rarely fit inside one tidy department. Maplewell's public team has included control engineers, data scientists, software builders and academic advisers alongside business-development specialists. His welcome to Dall'Anese on LinkedIn carried an unusually expansive destination for an engineering hire: an “equitable, decentralized, and decarbonized energy future.” The day-to-day work needed to get there is less lyrical. Someone has to reconcile a facility's meter data, the battery's state of charge, a tariff calendar, a forecast and a control command that equipment will accept. Irvin's role sits across that seam, part company builder and part systems architect.
“With the team support at Intel Liftoff Days, we felt comfortable pushing our boundaries and trying something risky.”Matthew Irvin, on experimenting with new forecasting models
Risk, with a baseline
In 2024, Maplewell joined Intel Liftoff, a startup program offering AI tools, cloud access and technical support. During an Intel-hosted hackathon, the team explored language-model methods for time-series forecasting, including Amazon's Chronos model. Irvin later said that without the program the team would not have tried the approach. Intel reported that the experiment improved Maplewell's forecasting accuracy by 25 percent compared with traditional methods.
The appeal was not artificial intelligence as decoration. Better forecasts let a battery choose a more useful hour. They help a controller anticipate a peak instead of reacting after the meter has recorded it. Intel's hardware work also addressed the less glamorous half of the equation: getting cloud-trained models onto resource-constrained devices at customer sites.
Irvin's public arguments keep returning to the same mismatch. Electricity demand is rising through electrification and data centers while generation retirements and long interconnection queues slow new supply. In a 2024 essay, he argued that distributed resources paired with intelligent energy management can arrive faster than large power plants and, when aggregated, provide meaningful grid capacity. “We have the tools; now, we need to use them,” he concluded.
The line fits his career, but the refrigerator tells the story better. A mechanical engineer accustomed to the heavy machinery of supply went looking for capacity in the schedule of ordinary equipment. He found a few kilowatts waiting inside a cold aisle, then connected that local bargain to batteries, buildings and a larger grid. The ambition is broad. The method remains pleasingly specific: find the peak, think ahead, move what can move, and check the meter afterward.