Explore a Decade of Diabetes Insights with Diabettech AI

Given the volume of articles on Diabettech.com over the past 10+ years of it’s existence, and the challenges navigating them, I’m excited to announce diabettech.ai, a new way to pull together the years of knowledge stored within Diabettech.

What is diabettech.ai?

Very simply, it’s an LLM interface to allow you to interrogate the website data and try and answer questions that might have been covered, whether that’s cures, approaches to life with diabetes or n=1 observational reports.

At its core, the system uses OpenAI ChatGPT 5o-mini to churn over the site and “understand” all the pages, then attempt to give you a result.

I’ve tried to ensure that the responses stick to Diabettech articles and comments, and if something comes up that isn’t within that dataset, it will say so rather than making something up. Now LLMs aren’t perfect, so even with the constraints in place, I can’t guarantee it will always behave itself.

A research tool rather than a conversational chatbot

In the initial phase, it’s set up more as a research tool than as a conversational chatbot. Am initial query is configured to throw up only a few article links, and if you want more you may need to hone questions to get them.

As we observe usage, we’ll be better able to tune this behaviour.

Given this is also a privately funded enterprise, with no external sponsorship, I’ve also set it up to try and limit the draw down on AI tokens (the currency of AI queries) for two reasons:

• Cost
• Resource usage

As a result, I try and ensure we cache queries where possible to reduce the amount of reuse of the underlying LLM model. If you’d like to help us cover the cost of using the underlying model, you can do so here.

What next?

Go ahead and try it out. Head over to diabettech.ai and see how you find it. It’s a beta trial at the moment that I hope is useful to some, and please provide feedback on your use in the comments.