This project is a mix of personal account and technical experiment, exploring metabolic science from the outside. In 2012 I radically changed my diet and have since been ignoring conventional nutritional advice - with great success and a noticeable positive effect on my health and wellbeing.
When I explain the background to people - what a “standard” diet does to insulin, and what insulin does in the body - I built the simulation on this site to make it tangible: you can see how different ways of eating affect blood glucose and insulin: Sometimes as spiky peaks, sometimes as shallow hills.
The simulation aims to be scientifically plausible, but is intended as an illustrative tool, not as an exact prediction for arbitrary scenarios. In particular, the displayed values are not (yet) calibrated against any real organism. Neither mine, nor yours, nor anyone else’s.
If you’re someone who can judge whether the numbers and their trajectories look realistic, I’d love to hear from you. That likely means: you wear a continuous glucose monitor, or you work in a related medical field. Either way, please get in touch.
Disclaimer
This project reflects my time spent as a nutrition nerd exploring approaches beyond the mainstream - particularly ketogenic and very low-carb diets. There is no medical advice here, no personal recommendations, only observations you can decide for yourself whether to look into further.
Feedback via the button in the bottom right, or any other way you can reach me.
AI Policy
These days it isn’t always obvious where content comes from - so here is how I use AI in building this site:
Articles/Blog
- The articles here are written by a human (me), typically in German, always based on my own experience, observations, and urge to share.
- Before publishing, I use an LLM(1) to check articles against a number of criteria - things like “is there a conflict with the German Heilmittelwerbegesetz?”, “does the argument hold together?” or “is study A characterised correctly?” - plus grammar and spelling, of course. Most of the changes that come out of this - structural and substantive ones in particular - I make myself, but I do adopt some of the suggested wordings(1) - sometimes verbatim, sometimes in substance.
- Studies in particular I look for using conventional search engines as well as with LLM(1) help. The studies themselves - unless I already know them - I read myself, to varying depths: at the end of the day I’m not a scientist, and scientific papers are written for a different audience. The abstract, the conclusions, a look at any conflicts of interest the authors may have, and an assessment of the methodology are always part of it, but LLMs help me understand and place the rest.
- Some outlines for new articles are supplemented with AI notes(1), which I read as topic suggestions. Those bullet-point notes occasionally give me ideas for entirely different articles, which then go through the same process.
- The translation into English is LLM-generated(3), although I edit it too, so that it uses vocabulary and grammar the way I would use them myself. Now and then the translation makes me notice an awkward formulation, structure or argument, which I then change in the German, have translated again, and edit again.
The simulator
Considerably more AI help went into the code of the simulator and the server-side components: I’m using this project to get to grips with technologies that weren’t previously part of my day job, so large parts of it are generated(3) - my own editing usually happened in the prompt, with the editor used for reading and understanding.
The simulator runs without any contact to any AI, or indeed to any system outside your browser and my own server. The analyses the simulation offers run deterministically, reproducibly and 100% without AI. Nothing you enter here is sent to an AI - neither for analysis nor for training. The AI involvement was in creating the analysis algorithms.
The texts of the analyses themselves are static content and - like the code - mostly generated(3).
When it came to tuning the simulation algorithm, an AI(3) helped connect the blood glucose curves I had with the underlying mechanisms, and to settle the parameters for the simulation. As much as that helped: there is a lot of human work in it, but without the scientific references and the rapid test runs for sanity-checking the values that came in via Claude, an algorithm like this would never have come about.
Illustrations
By far the most illustrations on this site I find on Unsplash. I credit third-party rights (and the licence to use them) with every image (the reminder to do so is part of the publishing workflow, which an AI initially watched over - by now it’s script-supported). As I write these lines there are two images that were created with AI assistance (one of them significantly reworked by me), out of 100 images at present - and I’d like to keep or improve that ratio in favour of human-made illustrations.
CMS
The CMS I use is Hugo, a static website generator. Elements embedded in articles are implemented with Hugo-native techniques (“shortcodes”), which I mostly have Claude generate. (Another technology that wasn’t previously part of my day job)
Impressum
Olaf Kock - Pinneberger Chaussee 28 - 22523 Hamburg (webmaster at olafkock dot de) - or via Mastodon @olafk@norden.social