I wanted to speak Dutch to my AI: faster than typing, richer in detail and closer to the way I actually think. The obvious market solution was a paid American transcription service that only partly understood me and processed my voice on infrastructure I did not control.

So I decided to build my own.

I am not a programmer. My first attempt failed. The second became a working Dutch speech-to-text application that runs locally on my laptop—without sending my voice to an external server.

In this twenty-minute first-person build story, I share what it really took: the hours, costs, false starts, technical compromises and the things that still do not work. Through a live offline demonstration, an abstract policy term becomes tangible: data sovereignty is not only about regulation or server location, but about having meaningful choices over the technologies on which we depend.

The audience leaves with three practical questions they can use to examine their own AI workflows: what data enters, where it is processed, and who ultimately remains in control.