ABOUT US
We did not build an AI and then go looking for manufacturing problems.
We spent years on the shop floor first, and that is how we learned which problems are worth solving with AI.
01, Where it started
RemoteNC's first product was not software. It was a machine tool.
In those years we lived on the shop floor: standing at the machine watching an operator change tools, waiting in the corner of an office for a customer to send a drawing back, keying in twenty-odd dimensions by hand before close of business for a rush job.
What we learned then was not technology. It was the real rhythm of this industry, the quote is due tomorrow morning, a revision must reach the floor the moment it changes, the inspection sheet has to be finished before the truck leaves. None of it is hard on its own. All of it has to be right, and there is a staggering amount of it.
Later we turned the potholes we had hit into an ERP and sold it to shops like ours. The process did get smoother, but the most labour-intensive part was never solved, somebody still had to read every drawing, one at a time. Only then could you quote, release a work order or produce an inspection sheet. And that somebody was usually the most experienced person in the building, exactly the person who should not be tied to it.
That is where CADASU began. We wanted to give that person back to the work only they can do.
Roots
We developed machine tools ourselves and hit the same walls, so we speak the same language as the floor. Nobody has to explain what a title block is, what a revision means, or why you cannot just loosen a tolerance.
Team
A research team out of world-class AI companies, alongside more than a decade of shop-floor experience. One side knows how to train the model; the other knows how that drawing actually gets used. They sit in the same room.
02, What we believe
01
Looking at the same drawing, sales sees a price, engineering sees feasibility, and QA sees whether it can be measured. If the people building the tool have never sat in those three chairs, what they build will be clever and unusable. Almost every spec we have written started as a complaint from the floor, not a paper.
02
In manufacturing, every number ends up in front of a customer. So we decided in the first version: every extracted dimension gets marked on the drawing, and every answer points to a document and a page. AI can do ninety per cent of the work, but the last ten per cent, the checking, has to stay with a person, and it has to be fast to check.
03
AI changes almost monthly. A system that takes a year to go live is running last year's technology on day one. So CADASU works out of the box: drop in a drawing you already have and look at the result. Once you know it helps, then we talk about integration, customisation and rolling it out. Do it the other way round and both sides usually regret it.
03, Milestones
From selling machines to selling software to training our own models, every step was pushed by the shop floor.
The company was founded around machine tools. The rules, the vocabulary and the rhythm of the shop floor were all learned then, and so was our running list of what hurts in this industry.
We turned the potholes we had hit into a system and started selling software instead of machines. From quoting to work orders to inventory and costing, that ERP still runs daily in customer plants, and it gave us first-hand knowledge of how manufacturing data is actually shaped.
We began developing models that read engineering drawings, what became CADASU AI.
With NVIDIA's technology and compute behind it, model training moved onto world-class infrastructure.
That same year CADASU appeared on the ecosystem partner wall shown in Jensen Huang's GTC keynote.
The funding went into model research and delivery capacity. Most of it goes to one thing: making the model read more accurately and making it easier to adopt.
04, Join us
Taiwan has tens of thousands of machine shops holding up a large part of the world's supply chain, and most of their drawings still sit in folders while quoting still depends on one person's memory. That is worth solving properly, once.
If you want to build something people actually use every day rather than something that demos well, we would like to talk.