The platform read paper and guessed. Now it reads 30 years of records and prices from them.

The archive

Every aircraft sale leaves a paper trail at the FAA. VREF had decades of it: scans, photocopies, forms filled in by hand in 1987. The value of the company was in that cabinet and nobody could search it.

OCR turned it into a database. Aircraft, owner, date, condition, price, thirty years deep. The first time anyone could ask "every Cessna 172 sold in Texas since 1995" and get an answer in a second.

The model

With the archive structured, a valuation stopped being an estimate. A machine-learning model trained on thirty years of comparable sales now produces the number, and weekly scraping of live listings keeps it priced to this week's market. The team that used to type valuations now reviews them.

The platform around it

• New Django backend, new React front end. The old one was older than some of its users.
• A public API that became a product line with its own subscribers.
• The pipeline that keeps the model fed: scrape, clean, retrain.

The number

Revenue up 67% the first year. My team at Horizon Dev has run the platform since. It brings in millions a year.

The hour-sized version

Get the data out of paper and PDFs. Let the model do the repetitive judgement. Keep it fed. In an hour we do that to one task. This is what it looks like after two years.