Hardware is where constraints become visible.
Eight years as the sole designer across five thermostat products, connected sensing devices that run from fixed-segment to touchscreen, DIY homeowner to hotel guest. Each had its own ways of failing, and all of them had to feel like one family. Keeping them coherent was never really a styling job; it came down to judgment, one call at a time.
Patents
US 12,608,066
Low Power Detection
Awarded · 2026
Patent pending
Segment display character set
Photo 1 of 5
A touchscreen is not a fixed-segment display.
Sensi Touch 1 was built to mirror the Sensi Smart, the Wi-Fi fixed-segment thermostat sitting next to it in the product line. For a first-generation touchscreen, that was a reasonable instinct. But treating a touchscreen like a fixed-segment display with more pixels left most of what the hardware could do on the table.
Touch 2 was the chance to fix that. The menu structure from Touch 1 was solid, and I had no interest in making users relearn navigation they already knew. What needed to change was the visual language, the interaction model, and the interface architecture underneath.
Before any design work, I went through Touch 1's strengths and weaknesses, mapped the competitive field, and interviewed utility partners, contractors, and homeowners. Each group ran into different problems, and hearing all three was the only way to tell which ones were worth solving.
One button, one job.
The change that mattered most was not a visual one. Every screen has a primary action, and on Touch 1 that action moved around or hid depending on where you were. I gave it a fixed home: a contextual action button in the same spot on every screen, always set to the most important next step. Users learn it once, and after that it holds everywhere. Nobody notices consistency like that until it breaks.
The architecture had to outlive the first release.
The less visible decision was building the interface architecture to carry features that were not scoped yet. Remote sensors and the Equipment Interface Module were nowhere on the roadmap when I laid out the framework, but I left room for them. When they arrived later, they slotted in without a rewrite. I treated the architecture as a design decision, not an engineering afterthought.
Physical and digital decisions were connected from the start.
I came in during the industrial design evaluation phase, running usability testing and feeding back on how each form-factor concept changed the UI. On a touchscreen the two sides are tied together at every level. Thickness moves the reach targets. Bezel width changes how much screen you feel you have. Screen size sets how dense an interaction can get before it feels cramped. I was the person sitting between those conversations, and it ran in both directions. The form factor shaped the interface, and what the interface needed turned into requirements I handed back to industrial design and engineering: where a control sat, how a component had to behave, what the hardware had to leave room for.
Touch 2 shipped at 0.77 inches, the thinnest smart thermostat in its category at launch. The interaction model later fed into consulting work with Trane's internal design team on their premium touchscreen thermostat, which was a good sign it held up outside our own product line.
The design problem was not the hardware. It was the trip.
The Equipment Interface Module bridges thermostats to HVAC equipment in installations where the existing wiring cannot carry everything the system needs. Plenty of older homes have four wires or fewer, and a heat pump can require six. Competitors covered that range with three separate modules; the Sensi EIM covered it with one.
The hard part was configuration, or really where the configuration had to happen.
The default assumption was that you would set the EIM up at the equipment. In a typical install, that means going back and forth between the thermostat and the unit, often through a cramped mechanical space or up and down stairs. Every one of those trips is time the contractor cannot bill for.
Move configuration to where the contractor already is.
So I moved it to the thermostat. The contractor pairs the two devices, sets the equipment type and location once, and the EIM takes that information and configures itself. No return trips to the unit.
It is not the sort of work that photographs well; the value is in what it takes off the job, the walking and the uncertainty and the lost time.
Contractors noticed the part that mattered.
In training sessions after launch, two separate distributor groups brought up the pairing process on their own — being able to configure from one place, without moving between the thermostat and the equipment. The launch webinar drew more than 180 contractors.
Contractors tend to be vocal when something does not work, so silence carries information too. The pairing flow generated no iteration requests after launch. It did what it was supposed to.
Thirty-two segments, three buttons, no room to hide.
Sensi Lite was the most constrained product I have worked on: thirty-two display segments and three buttons — up, down, and menu/action. The brief was a complete thermostat experience for a budget-tier product that still had to feel considered.
With that little to work with, most of the design is deciding what to cut. A segment that is not doing real work, a button press that goes nowhere, a setting hidden out of habit rather than reason — none of it survives.
Navigation stayed flat because hierarchy was too expensive.
Three buttons make hierarchy expensive. It is the same problem as designing for a remote: every state has to be reachable through a tiny, fixed set of inputs. So I kept navigation flat and cyclical. Rather than drilling into nested menus, users loop through the settings, and the menu/action button does different jobs depending on how it is pressed. A normal press moves through the flow. A long press opens homeowner settings. A second long press reaches contractor configuration.
That last layer earns its obscurity. Contractor settings can break an HVAC configuration if a homeowner wanders into them, so the long-press-to-long-press gesture keeps them within reach for installers and out of the way for everyone else. Owners who want to get there still can.
With flat navigation, order becomes logic.
When navigation is flat, the order of the settings does real work; it is not just a filing system. Some settings depend on earlier ones — set up a heat pump, and reversing-valve direction has to come right after. So I ordered everything by those dependencies and put the most common equipment configuration first, which shaves time off the install.
Testing had to include the whole product.
In usability testing, 23% of users stalled at the three-dot menu button because it did not read as something you could press, and auto mode made 30% hesitate — a sign the conditional sequencing was pulling its weight. Icon testing ran through five variants and settled on a square inside a circle, lifted from the media stop symbol; that one carried into later products. I also sat in on physical installs, watching people tuck wires, snap the unit onto its baseplate, and fight with the clips, and I turned what I saw into hardware requirements, from clip resistance and baseplate fit to where the controls sat. On a product this small, UI and physical design stop being separate problems.
When the display goes dark, the system still has to communicate.
Sensi Lite runs on power stolen from the HVAC system, and when that power drops too low the display cannot run. That raises an odd question: how does the device tell you what it is doing once the screen you would normally check has gone dark? I was part of the team that worked out the answer — what shuts down, in what order, and how the device reports its state when it cannot show anything itself. What we landed on was a prioritized shutdown sequence plus a notification flow that pushes device state through the cloud to the mobile app, so the phone stands in for the display when the thermostat cannot. The low-power detection method behind that shutdown sequence is the subject of US Patent 12,608,066, awarded April 2026.
The decisions I lost changed how I work.
Two of my positions did not survive stakeholder review, and the post-launch data proved both of them right. I wanted fan-circulation runtime surfaced higher for homeowners; it got pushed down in favor of a mobile-first vision and turned into a recurring support call. I argued for moving the menu/action button to the far side of the display to cut down on accidental capacitive taps; hardware engineering preferred keeping it in line, and those taps became a post-launch problem that needed a firmware patch.
I keep those two around for one reason: they taught me to write down the reasoning behind every test. If you are going to lose an argument, lose it on the record — then the evidence is already there when the data comes in and settles it.
Designing for guests who will never be onboarded.
The Verdant VX4 and Line Voltage thermostats serve the hospitality market, which is a different world from residential. Hotel guests speak different languages, bring wildly different comfort with technology, and touch the thermostat once instead of living with it for years. There is no onboarding, and no help text a guest will stop to read. The interface has to land immediately, with anyone, or it has failed. This is lean-back design in the literal sense: read from across the room, understood at a glance, operated without instruction.
Text was not a reliable tool.
The Line Voltage model for the European market pushed that further. Text could not stretch across the range of languages the product had to serve, so I dropped labels and built the interface out of icons instead, a visual language tied entirely to function. An icon either reads or it does not, and there is no caption underneath to bail it out.
The character set had to work from across a hotel room.
Standard 7-segment displays are drawn for manufacturing, not for reading, and reading a temperature from across a room is a different problem than reading it up close. I started by studying existing segment forms, both standard and non-standard, then drew ideal numerals in Lato, the same typeface we already used in Sensi Touch 2. On transparency paper, I overlaid every digit to find the points and cut lines they shared, then pulled the segment count down by spotting segments that always fire together and treating them as one. That cost me some accuracy on the rarer letterforms, like W and Q, but testing showed they stayed legible in the places they actually turned up.
The testing had to match the room.
The finalists went to A/B testing on PCB prototypes from our supplier, in a room big enough to stand in for a hotel: 15 to 20 feet of viewing distance, with participants who had never seen the product. We changed the lighting, put random text on the displays, and measured both how accurately people read them and which versions they simply preferred. The character set that came out of it is the subject of a pending US patent, with the novel claim sitting where the segment geometry meets the reduction method.
Knowledge should not stay trapped in one product line.
The wireless sensor protocol we built for Verdant's hardware later turned up in Sensi's room-sensor network. Not because the two products share an ecosystem, but because the underlying problem was identical and the work already existed. When the same problem surfaces in two places, expertise from one is worth reaching for, and walling it off by product line carries a real cost.