FireGEO
Their analysis took 90 seconds. The page had to feel fast anyway.
The startup had done the hard part already. Two years of satellite imagery processed into wildfire perimeters, spread velocity and atmospheric conditions, and a data science team who knew what all of it meant. What they had never built was anything a customer could actually buy.
They’d tried building a dashboard themselves and got stuck on billing.
Two things made this awkward. Their analysis took 30 to 90 seconds per query, and nobody waits 90 seconds looking at a spinner. And if you charge per query, you have to decide what happens when a query dies at second 70.
FireGEO is Mendable’s project. I built the product layer on it.
Results stream as they arrive. The page shows something within about two seconds and fills in from there. The analysis still takes 90 seconds. It stops feeling like it.
Credits are reserved, then committed or refunded. Charge at the start, settle at the end, refund if it fails. Stripe webhooks reconcile the balance so the two systems can’t drift apart.
Results are stored as typed JSONB through Drizzle. Perimeter polygons, spread vectors and risk scores don’t share a shape, and forcing them into one table was going to hurt later.
They asked for six weeks. It took seven, and they had a paying customer at the end of it.