Small, on purpose

A vendor comes to a workshop with a demo. Their AI has read every manual, every forum and every fault report, in every language there is. It answers anything you ask it. It is genuinely impressive.

Someone at the back asks what happens when the internet drops.

It stops answering. That is the honest reply. The system lives in a datacentre a long way away, reached over a line the workshop does not own and cannot fix. In most workshops that line goes down at least once a week, usually on the day the yard is busiest.

That trade never makes it onto the slide. It is also the thing that decides whether the tool is any use at four o'clock on a Thursday.

Why bigger is not better here

There is an assumption underneath most AI marketing: a model that knows more is a better model. For a general assistant, fair enough. For a workshop, it is the wrong measure.

You do not service every machine ever built. You service the ones on your floor — a dozen models, perhaps two dozen — and you hold a repair history that belongs to those and to nothing else. A model built to know a little about millions of machines spends nearly all of itself on machines you will never touch.

That is not a free choice. It is exactly why such a model needs a datacentre to live in, and a connection you cannot guarantee in order to reach it.

So we ask a different question. Not "how much does it know", but "does it know your machines, and is it there when you ask".

How small actually works

VEDA is trained on the machines a workshop or an OEM actually services: the signals those units put out, the repair history they have built up, the notes the engineers on that floor have written. It knows less than a general model about the world, and a great deal more about your fleet.

Being that size, it fits on one ordinary computer you can buy, install and cool in a corner of the building. Three things follow from that, and they are the whole reason we do it:

• It is quick. The question never leaves the building, so the answer comes back at about the speed you can read it.
• It is there. The line to the outside world stops being part of the diagnosis. A dropped connection becomes an office problem, not a workshop one.
• Your history stays yours. The signals, the job cards and every verdict a technician gives stay where they were generated.

That last one matters most if you are an OEM watching over many workshops rather than one shop with one bay. The record that makes VEDA better is the same record you would never want leaving the site — and here it never has to, because it was never sent anywhere to make VEDA work in the first place.

Small, on purpose

Small here is not a compromise made to save money, though it does. It is not a trick to go faster, though it is faster. It is the size the job asks for: reasoning about a set of machines one company owns, using a record only that company holds.

A bigger model would not understand those machines any better. It would only be harder to keep next to them.

And it ends where everything we build ends. The machine proposes, a person decides, and the record stays in your building.

VarahiEdge · the machine cell · drag to turn