Researcher AI & Process Automation

QDid you always like computers?

I have liked computers since I was a kid, and I used to buy parts and assemble them myself. That interest carried into college, where I double-majored in computer science and mathematics.

QSo your major must have helped the most in this job.

Honestly, I think which school or department you graduated from matters much less than it used to. What matters now is what problems you have solved, what questions you asked and how you got to the answer, and what you have actually built. Inside the company, too, people are judged on skill and results, not on their education or credentials. And the starting point for all of that is curiosity about AI.

QYou probably did not know much about piping. How do you learn?

I say plainly when I do not know something, and ask the person who knows that field best. On site there are people who have worked with piping and processes for a long time. Listening to them all the way through and learning from them is half my job. I do not try to know everything myself. I leave specialist knowledge to the specialists, and focus on making that knowledge work properly inside our tools.

QAs a researcher, do you build AI models all day?

Surprisingly, I do not spend much time building AI models themselves. Most of my time goes into building the apps and websites that collect the data an AI will need. Semiconductor piping jobsite data does not exist anywhere on the internet, so gathering it ourselves is the real starting point of AI development.

QWhat does building apps and websites have to do with AI?

AX only works when DX is done well first. Honestly, much of the company’s work is still not organized digitally, so we are moving the whole company’s work processes onto digital tools. But it is a little different from typical DX. We collect data with AI in mind from the start, and connect the data we need to build our own AI.

QWhat kind of app do you build to collect data?

A representative example is the 3D Pipe Editor. It moved bend-value calculations and interference checks that used to be done by hand onto the screen, so the person doing the work is faster and more accurate, and the process leaves data behind. I cannot share exact figures since they are confidential, but when we first saw the results, the efficiency gain surprised everyone.

QWhich work do you change first?

We start with the points where the people doing the work struggle most. We use every tool ourselves first—“dogfooding”—to filter out what is inconvenient before handing it to the team. We also run AI training for anyone interested, so they can improve their own work themselves.

QHow do you solve problems?

I keep to an order. First I question the requirements, then boldly delete any step that can go. I simplify what remains, speed it up, and automate last. The most common mistake is automating work that should not exist. And after one lap, I start again from the beginning.

QWhat AI do you use?

We use the best models available at the time. As of September 28, 2026, I develop with Claude Opus 5.5 and GPT-6 Astra, and we are testing Jev for tasks that need fast judgment. Models keep changing, but the principle of always using the best one available stays the same. There is also effectively no limit on usage for work, which is the biggest advantage of this role—you can experiment freely.