Before designing features, I needed to understand how homeowners actually hold knowledge about their houses, and where that knowledge breaks down. Domis has to meet people in that messy middle, not ask them to become facility managers overnight.

Domis only works if it knows your home, but listing every space, appliance, and nuance by hand is painful.

When people tried to set up a home profile, many realized they didn’t know their own homes. Setting up Domis also became the act of learning it.

So to make setup feel as rewarding and informative as possible, I designed Domis around one principle:

Extract the maximum value from the smallest action the user is willing to take.

This is what an owner who understands their house looks like in software. A filled property. Tasks tracked with tagged appliances. Tasks pulled from the report.

The first thing anyone does in Domis is type their address. That is the entire ask, and it is roughly the maximum a new user will tolerate before deciding whether this is worth their time.

So I designed an address intelligence feature that maximizes the value of a user provided home address. Type it once, and Domis researches the house, then puts what it finds into the same form the owner would have filled by hand.

The hard part is not finding data. It is deciding what the owner should see when search is unsure. The same address can return different facts on different runs.

Rather than forcing a confident answer, Domis does 3 searches, cross references them, and only shows what sources agree on, leaves blanks when they do not, and never locks the owner out of a field.

Logging your appliances is one of the first useful things to do in Domis, since it lets them be tagged in tasks and in conversations with Pros.

Walking around photographing them turns out to be mildly satisfying. Typing in the model number and other minute details does not.

The model number was not the user goal. It was a key to the things owners actually need later: manuals, warranty, support, parts, and common failures.

I asked my Google Gemini app about appliance labels to see what it could give me.

With the Gemini test as proof of concept, I moved on to designing the feature.

V1 stopped at reading the plate. For v2 I pushed the same scan further: support pages, warranty links, the things an owner actually needs when something breaks.

A key part of my research was around meeting the user where they were, and I went looking for what information homeowners already possess.

Every homeowner has an inspection report. It is also long, repetitive, and almost entirely unread.

However, no one cares about the inspection report. Our interviews explained why. People had been in their homes long enough that the report no longer described the house they were living in.

That changed what the feature should be. Not a diagnostic that generates a to-do list, which would be wrong on arrival, but a way to understand the house.

Systems. AI output and human input share the same editable forms. Mobile and web sit on one design system I own. Different ingestion paths share one contract: fill what you can, never lock the owner out.

Trust. AI is wrong sometimes. We show disagreement instead of a confident wrong answer, and treat the report as understanding, not an automatic chore list.

Collaboration. As founding PD I set phases with leadership and eng: what ships when, and what the AI is allowed to claim. I define and prototype the AI surfaces; engineering owns the pipeline under that contract.

Domis is recently launched and rapidly updating.

The lens has not changed. Find the smallest thing the user is willing to do, and return as much as possible for it.