Small tools, useful experiments.

My first practical Python project was a local availability monitor for climbing Pico in the Azores.

It regularly checked for open booking slots and filtered them by the time window I needed. As soon as two suitable places became available, it sent me an email.

It was a small tool with a very specific purpose, but it showed me what a few lines of code and some automation could do.

Later projects combined Python with language models.

One example was a review tool for environmental claims. It used structured checking rules and an AI connection to identify potentially problematic wording and suggest clearer alternatives.

Some of these experiments grew out of professional challenges, so I keep the organizations, data and internal processes behind them intentionally private.

readdit began as a vibe-coding Advent calendar that I received as a gift. At the start, I had no idea what I would be building or where the daily tasks would eventually lead.

Each day, I opened a small paper note with one manageable task: create the first structure, connect an API, add a feature or fix something that had stopped working.

A custom GPT called Advents-Fee accompanied the project. It translated the daily tasks into practical steps, explained unfamiliar parts and helped me keep moving whenever the code did not behave as expected.

By Christmas, the individual pieces had grown into a working website. readdit combined Open Library data with community signals from Reddit to help discover books based on selected preferences and personal reading taste. For a short time, the project was publicly available at readdit.de.

More recently, I have been working with specialised AI agents.

The interesting part is not simply creating another chatbot. It is defining what an agent should know, which sources it may use, how it should approach a task, where it needs to be strict and how its results should be structured.

Much of the work happens in the instructions, knowledge sources, review logic and repeated testing behind the visible conversation.

For me, vibe coding means building software through an ongoing conversation with AI.

I describe the problem, create a first prototype, test it against reality and gradually refine both the code and my own understanding of it.

The first generated version is rarely the finished one. The useful part begins when something does not work as expected.