Ugong: turning a strange engine noise into evidence

The problem

A vehicle starts making a noise. The owner can only describe it in words, the noise never happens at the workshop, and parts get replaced that were fine.

The plan I wrote

A fixed recording routine — the phone in named positions, the engine held at stated speeds — so two recordings can really be compared. The sound is analyzed against engine rotation instead of time, which turns a blurry smear into a sharp signal.

The app shows where the extra sound energy sits and how sure it is, compares it with the same vehicle's earlier recording, and hands the mechanic the evidence. It never names a part to replace, and it never records while the vehicle is moving.

The build

AI coding agents built it from my written specification. My work was the specification, the tests it had to pass, and the review.

The proof

The first check was wrong: an agent marked the app ready after only opening it — the sound analysis had never run. Making it run for real exposed three bugs: the results were too big to save; the first fix then threw away the very peak it was meant to find; and the summary described a different problem from the one it measured. All three were fixed, and an automated test was added.

Then a test that could fail, on the final release build: a 400 Hz tone read as a 3× pattern and an 800 Hz tone as a 5× pattern, both with high confidence. The result changes when the sound changes — so it is measuring, not guessing.

The launch

Checked on real devices — an Infinix phone and a Galaxy Tab S3 — rebuilt for Google Play's newest Android requirement before its deadline, and live on Google Play with its store listing, screenshots and privacy page.