How it works

The questionnaire is the main tool

Every diagnostician starts by asking: when did the noise appear, what does it sound like, does it depend on speed, engine revs, braking, cornering. Your answers rule out whole families of causes — that is worth more than any algorithm running on a poor recording. So the questionnaire comes first in Pro-Stuk, and its logic is a decision tree: each answer leads to the next, narrower question.

What happens to the recording

The 15–30 second recording goes to the server. First it is handled by ordinary maths, no neural networks: the spectrum of the sound, its colour (a tonal squeal or broadband noise), the rhythm of the knocks and their rate, an estimate of engine speed from the low-frequency band. Every feature is marked for reliability: if the recording is quiet or noisy, the features are honestly flagged as unreliable.

Then a language model puts it all together: your answers, the features of the recording, the audio itself and the car — make, year, mileage and the known weak spots of that model. The result is 2–4 likely causes with percentages, an urgency level and pointers for the garage.

Why the result is a probability

Different faults make the same noise: the hum of a wheel bearing is easy to mistake for tyre noise, and a knock from the anti-roll bar links for something far more serious in the suspension. Telling them apart for certain takes an inspection on a lift. So Pro-Stuk does not give a diagnosis and does not promise accuracy — it spreads the probabilities honestly and tells you what to check first.

A good recording improves the estimate, but it does not replace a mechanic. Treat the report as a second opinion before you go in: it makes the conversation specific, and makes it harder to sell you work you do not need.

Data

The audio is analysed on the server and is not kept afterwards. There are no accounts, no analytics and no advertising trackers. More detail in the privacy policy.