Who runs this, and how to reach me
Grounded AI Practice is an independent research and practical capability project based in the United Kingdom. It researches how AI skills are being taught and measured in the UK, publishes what it finds with the sources attached, and builds practical tools and learning material for the people that provision currently reaches least: individuals, and small organisations without a training department behind them.
I read everything that comes in. If you are writing about the research, about the workstation, or to tell me something on this site is wrong, all three are welcome.
Business details
- Trading name
- Grounded AI Practice
- Nature of business
- Independent AI research, practical AI capability development, and learning material for individuals and small organisations.
- Location
- United Kingdom
- Established
- 2026
- Website
- groundedaipractice.co.uk
- Repository
- github.com/OHPS-Stack/grounded-ai-practice
What this practice does
Three strands, at three different stages. I would rather say which is which than present all of it as though it were on a shelf ready to buy.
Research and written work
Sourced research on the UK's AI skills programme, its published figures and what they can and cannot show. Reports, the full evidence log, and the freedom of information requests behind them are all public.
Learning units
Short practical units on using AI well, built to be finished in a sitting and tested with real learners before being scaled. Drafts are public; none has completed its trial.
On-premises AI systems
An AI workstation that runs on a small organisation's own hardware, carries workflows built for its actual work, and teaches its own use. A working hypothesis being proven on the project's own machine first.
If you represent a small organisation that would find an on-premises system useful, I am interested in hearing what you would actually want it to do, whether or not anything comes of it. The demand side of this is the part I have the least evidence on, and I would rather ask than assume.
How I work, stated up front
I use AI assistants throughout this project, for research legwork, drafting and building tools, and I review what they produce before it goes anywhere. That is disclosed on every piece of work rather than mentioned once and forgotten, because working with these tools without handing them your judgement is the skill the project exists to teach.
Every factual claim published here traces to a source you can go and read. When something turns out to be wrong, the correction goes in the log where the original was, and the original stays visible. The method page sets out the four rules in full, and the logs in the repository show them in use, including the times they caught me out.