Web and iOS · Vision model
RentGrade
Grade an apartment before you sign the lease.
Our product, on the App Store
Renters photograph the apartment on the walkthrough, and RentGrade turns those photos into a graded condition report with every problem marked on the picture that shows it. We built the app, the backend, and the iOS release, and it went from Brad’s idea to the App Store in a month.
What it does
A renter photographs each room on the walkthrough. RentGrade returns a graded condition report with every problem marked on the photo that shows it, a repair estimate to take to the landlord, and a lease you can ask questions about in plain English.
How it works
A vision model grades the whole photo set in one pass at temperature zero, then sweeps each photo for anything the first pass missed. The model answers with a coarse grid cell rather than a pixel, so a second pass runs on the cropped region and turns that cell into a pin that lands on the defect. Findings are merged so the same crack is not reported twice.
The hard parts
Getting a pin to land on the damage instead of the sky. Keeping a long generation alive on an iPhone, where the web view kills a request at thirty seconds and the server allows sixty. Moving every database query behind the server so the app never holds a credential.
Built with
React 19 and Vite, twelve Postgres tables on Neon, S3 for photos, Vercel functions, and a Capacitor iOS build with its own Swift camera and PDF code. 168 tests.
Visit rentgrade.ai (opens in new tab)
