Proof Before Pay
A voice assistant add-on that checks the contractor’s license and pays only for home repair work that has proof.
The live demo is a simulator built for a computer screen.
Read the code on GitHubInvoice check · Bright Current Electric
demo dataPay now
$1,050.00
- 12 standard outlets$360photo 10:15 AM
- 2 GFCI outlets$190photo 10:42 AM
- 6 light switches$150photo 10:58 AM
- 3 bedroom smoke detectors$210photo 11:12 AM
- Master bath fan circuit$140photo 11:33 AM
Waits
$370.00
- Hallway smoke detector$70no photo
- Electrical inspection$300not in the agreed scope
From the demo. All businesses, people and places in the demo are fictional. No real money moves.
Four checks, and every check names its source
-
1
License
Looks up each pro’s license number in Florida’s public license registry and leaves out anyone whose number isn’t on record.
Source: the registry and its current-as-of date -
2
Scope
Before booking, it reads back what you are buying item by item, and what will prove each item.
Source: the agreed scope -
3
Visit
With the pro’s consent, the front door camera records only arrival and departure for that job.
Source: door camera times -
4
Payment
Fixed rules (not an AI model) match every line against the scope, the visit times and the photos. Only proven items go to checkout.
Source: the photo
The working prototype
Screens from the running prototype, in the Echo Show simulator I built. Fictional data.
The public numbers
- 2,830complaints about unlicensed construction contracting, Florida DBPR, fiscal year 2024-25
- $4.35 millionpaid by Florida in 2025 to homeowners harmed by licensed contractors
- 81,925FTC reports in home improvement, repair and solar, 2024
- $1,405median reported loss, home improvement scams, BBB Scam Tracker 2025
I ran the rules over my own record
141engagements
15booked jobs (13 after we closed on the house)
5went wrong
All 5 would have hit at least one check.
3 of the 5 would have had every problem caught.
Up to $1,219 I had already paid on them would have waited for proof.
The 13 jobs I booked after we closed on our first house in July 2026
- Marked done before the work was finished
- Marked done before the work was finished
- Crew damaged the dryer duct
- Missed the agreed arrival window
- Missed the agreed arrival window
- Marked done before the work was finished2
- Crew damaged the dryer duct1
- Missed the agreed arrival window2
In two of those jobs I had already paid before I could check the work.
On my history the visit check would have flagged 4 of the 5 bad jobs, payment only after proof 3, and the license check 1.
What is built, and what isn’t
Built and tested
- Server with 8 tools and six screen cards
- Florida license lookup, live for one demo number
- Invoice check with fixed rules
- Sandbox marketplace and checkout, no real money moves
- Classic Alexa skill on the same server, run in the Alexa Developer Console simulator
- 75 automated checks pass on a fresh copy of the code
- Public code under the MIT license
Not built yet
- Real camera events (simulated today)
- Real payments
- Use by other families
- Alexa+ opens add-on tools only to select partners, so the Alexa+ side runs in a simulator I built
What I need next
Partners
more than one state’s license registry and a home services marketplace
A pilot with real families
through housing counselors and nonprofits
Legal advice
holding part of a payment, door camera records, contractor licensing. In Florida: the construction lien law (Chapter 713) and the rules on contractor deposits (s. 489.126)
Measurement
what changes in a family’s finances
Many families don’t own an Echo Show. The checks don’t depend on Alexa: they run on an ordinary server, and the demo already runs in a web browser. So a pilot would start with a phone web page or text messages, and Alexa+ would be one way in, not the only one.
Who is building it
Mikhail Oskola
Founder, RetailBox Automation, Inc. · Orlando, Florida
I am the user this was built for. We bought our first house in Orlando in July 2026.
470+ jobs on Upwork · 100% Job Success · Top Rated Plus · 480+ clients
Winner, AWS Agents for Humans Hackathon (Everyday Agents track), project StepSpotter


