AI projects

Tools I build with AI

I use AI as a development partner to build the tools I need myself. Not as a shortcut, but as a method: a clear spec, measurable requirements and testing on real data before anything goes into use.

01 Projects

Selected projects

The transcription app showing a project meeting with three speakers

2026

Transcription and dictation

Two Mac apps that turn Danish and English speech into text with speakers, timestamps and editing. Everything runs locally on a MacBook Air and replaces a paid cloud subscription.

Claude Code
Python
Swift
MLX
Whisper
pyannote

Read the case

02 Method

How I work with AI

A Direction

I steer, the AI builds

I set the goal, the constraints and the acceptance criteria. The AI writes the code in small, testable steps, and nothing goes into use until it has been tested.

B Measurement

Measure, don't guess

Error rates, timings and memory use are logged on every test. When two solutions compete, the numbers and a blind listening test decide.

C Judgement

Data decides, even when the answer is no

Several ideas that sounded good were dropped because the test said otherwise. Knowing when AI is not the answer is part of the craft.