# My first AI was built with a soldering iron

Published 12 May 2026

[Originally posted on LinkedIn](https://www.linkedin.com/feed/update/urn:li:activity:7459938935289569280/)

![Illustration of a boy in a striped T-shirt soldering a circuit board next to a home computer and loose keys, with a note on the wall: "around 1990, somewhere in Ukraine".](https://ovchyn.me/api/media/file/my-first-ai-was-built-with-a-soldering-iron-1.jpg?prefix=production)

Before Python, before GPUs **my first AI was built with a soldering iron**. 🔧

In the 1990s, a **ZX Spectrum** — an 8-bit home computer — showed up in our house. ❤️ And that was it. Love! One that, honestly, has lasted to this day.

Back then it wasn't just a computer, it was a construction kit. You had to solder the keyboard — every single key. For real. Assemble the boards, put everything into a case.

The case was awful, by the way. So, I built my own — wooden. Moved the keyboard out separately. Console plus keyboard, my own vision, at age 8.

After that I started programming. Small things, experiments. I wrote a couple of programs. I don't even remember what the first one was about, but there was an*algorithm*, and I made it work. That was enough for me back then.

Around 2000, things got really interesting. I started reading a paper magazine called Computerra. It explained early AI research papers in plain language — the same ideas that later landed Hopfield and Hinton the 2024 Nobel in Physics. 🏆

What fascinated me was the simplicity and beauty of how it actually works.

Those papers suggested a simple idea: *don't brute-force everything — cut off dead ends early*.

I applied this to formal mathematical logic. Instead of hours or days, I got the Pythagorean theorem down to 30 minutes on an 80286. Just by teaching it to notice earlier: this makes no sense, we've already been here, this leads to a loop. 🔄

That was my first AI-related program. It made me a member of the Junior Academy of Sciences of Ukraine, in 11th grade.

And I kept poking at it. Later I came up with a self-hosted GenAI system for banking document generation — basically a ChatGPT for German banking, built before ChatGPT existed. Now there's **TrueBDD**, a small tool where documentation is the source of truth, so code can be deleted along the way.

AI still interests me — what's inside it, why it works, and where its real limits are, and how it will change the way we build software forever.