The fault nobody can place

A truck comes into the workshop with a fault nobody can place.

One technician says it is the injector. Another is sure it is a sensor. The third has seen this before and cannot remember how it ended. Two hours go by. The bay is still occupied, the customer is still waiting, and the truck has not moved.

Here is the part that should annoy you. The answer was in the building the whole time.

It was in the signals the truck puts out on its own wiring, every second it runs. It was in the repair history of the same model that came through last year. It was in a note the senior engineer wrote on a job card, in his own shorthand, that nobody has opened since.

Nothing is missing. It is scattered, and none of it is in a form you can ask a question of while standing next to a truck with a torch in your hand.

What we think the real problem is

Most people call this an AI problem. We think it is a memory problem.

A workshop already knows how its machines fail. That knowledge sits in three places: in the machines, in the paperwork, and in the heads of the people who have been there longest. The first two are unreadable in practice. The third walks out of the gate at the end of every shift, and one day does not come back.

So the job is not to invent a cleverer mechanic. It is to write down what the workshop already knows, in a form a machine can read — without asking anyone to fill in a form, because nobody fills in forms.

That idea is underneath VEDA, and underneath everything else we build.

How it works, roughly

VEDA is a small language model that runs on one computer inside your workshop. Not in a datacentre. Not on our servers. In the corner of your building, on hardware you own.

It reads three things:

• What the machine is saying. A modern vehicle talks constantly over its own internal wiring. VEDA listens to that traffic.
• What happened last time. Previous jobs on the same model — what was replaced, and whether it fixed anything.
• What your engineers wrote. The job cards, the notes, the shorthand nobody else can read.

Then it answers with a ranked list rather than a verdict. Check this first. Then this. And here is why.

Two things in that sentence do most of the work.

Ranked, with reasons. A tool that says "it is the injector" and is wrong twice is a tool the workshop learns to ignore. A tool that says "start here, because these three things are true" survives being wrong, because the technician can read the reasoning and argue with it.

The argument is the point. When the technician says no, that answer goes into the record — and it is the only thing VEDA learns from. Not forums. Not the manual. The verdicts of the people who actually opened the machine.

What you have after a year

On day one, a technician gets a better place to start than a guess. That on its own is worth the exercise.

What builds up is the more interesting part. After a year the workshop has a written record of its own reasoning: what went wrong, what was tried, what it turned out to be, and who confirmed it. When someone retires, that stays. When someone new starts, they are not beginning at zero.

All of it stays on the box in your building — the signals, the job cards, the photographs, the verdicts. That is not a setting we offer. It is why the system is shaped this way. A record of how your fleet fails is worth something, and it should be worth it to you.

And every step ends with a person. VEDA proposes; the technician decides. That is the part we are not trying to automate.

VEDA · drag to turn