Product · First trade: garage doors

AI diagnostics and dispatch for home services

Working build · private Home servicesAI2026

The homeowner talks and sends photos. The shop gets a diagnosis, the right tech, the parts to load and the truck logistics, before anyone drives.

ProofBuilt on 204 real cases. Cleared 11 of 11 structural acceptance gates.

The home page of a demo garage-door company, headed "Your door quit. We show up knowing why.", with one button, "Tell us what happened"

Copperline is a fictional company I built for the working demo. Its reviews and phone number are invented.

My role

Conceived it. Did the research, design and architecture. Built it, the demo company’s site included.

Results

Diagnosis, tech match, parts and truck logistics in one brief, built the moment the homeowner finishes.
Estimates job value and revenue per truck-hour, including drive time.
A corpus of 204 real cases, mined from 504 source candidates.

What it is

A home-services dispatcher takes a call from someone who can’t describe what’s broken. The tech arrives without the part. Second trip. Lost margin.

This fixes the first five minutes. The homeowner taps one button on the shop’s site, says what’s wrong in her own words, and sends a couple of photos. The system reads it against hundreds of real cases and hands the shop a brief: what’s broken, who to send, what to load, roughly what it’s worth, and how sure it is.

I built it for garage-door companies first. I’d never worked in the trade. That was the point. Different industry, same approach: start with the problem, work the solution, then build the system.

Watch it work

An edited walkthrough of a homeowner’s intake on a phone. Working demo, fictional company and customer.

Screens

A phone recording a homeowner’s spoken answer to "In your own words, what’s the door doing right now?" — She talks. It listens. No app, no form to fight.

A phone screen that repeats back what it heard, "a loud bang or snap, and now it won’t open", and offers closer options — It reads back what it heard and lets her correct it with a tap.

A phone screen asking for a photo of the spring above the door, because it decides which parts the technician brings — It asks for the one photo that decides which parts go on the truck.

A confirmation screen telling the homeowner the shop will reach out already knowing her door — Done. The next call she gets is from someone who already knows the problem.

A dispatch brief marked "Needs evidence", job value and revenue per truck-hour held until one more photo arrives, with a safety warning — When the evidence isn’t in, it says so. Price held, one photo requested, no truck sent on a guess.

The shop’s dispatch brief, with job value, revenue per truck-hour, time on site, drive time, a safety warning and the diagnosis, a broken torsion spring — What the shop sees. The diagnosis, the job’s estimated value, time on site, the drive, and a safety warning.

The dispatch packet, listing the parts to load and the technician skill needed — The dispatch packet. Parts to load and the tech to send.

More proof

It flags safety first. A door under tension gets a warning before anything else.
It says what it doesn’t know. With no photos, it holds the job value instead of guessing.
The recognition layer is grounded in how homeowners describe a broken door.
It sits beside the shop’s own website and phones. Nothing gets ripped out.
It runs on the same foundation as Wavelength, and it’s shaped per trade.

Related work

All products