JAY MAUDE
Navigation
Home,
Work,
Contact
Creative Director
Based in Oakland, CA
About 250,000 lines of Python, more than 30 checks per file
Runs locally, built with AI coding tools
The Project
Mira is a QC app for checking video deliverables before they go out the door, which I designed and built with AI, using Claude Code. It’s around 250,000 lines of Python, runs locally, and makes more than 30 checks on every file.
The Challenge
A frozen frame and a deliberate hold look identical to a computer, and loudness targets change depending on where a film is going. Every check meant working out what a real error looks like in human terms, then how a computer can find it deterministically.
The Approach
It started on the first Lightroom campaign film: so many deliverables, and some very late nights with tired eyes, that checking them by hand was a huge lift. I built a small tool to help, and it kept growing. I developed it the way I’d run any project, in iterations: build a version, run it on a real QC pass, fix what it got wrong, go again.
Most deliverables differ only in format, aspect ratio or call to action, so Mira is built around watching one of each. Once you’ve approved the master, it checks every other version against it pixel for pixel, reads the supers, CTAs and legal copy, and matches each end card against the deliverables tracker. A likely problem shows up amber, a confirmed one red, and anything Mira couldn’t check stays grey, so a file nothing looked at never passes for one that did.
The Impact
It can’t hallucinate a problem, because the same file gets the same result every time. It already saves hours of scrubbing through near-identical files, and I’d like to bring it to market soon.