Three things get published here, and nothing else: concrete AI use cases I have implemented or evaluated, news that changes what is practical for a business to run, and blueprints — the actual configuration for running models on your own hardware, written so you can reproduce it.
The full configuration for serving an open-weight model on your own hardware — hardware sizing, the OpenAI-compatible endpoint, environment variables, and the failure modes nobody documents.
Read the blueprint →Five processes that reliably return the investment, three that reliably do not, and the test I apply before recommending either.
Read the article →Why constructionism and cognitive science can turn complex computer systems into tangible, engaging learning experiences.
Read the article →A short monthly note: model releases, licence changes and pricing moves — filtered down to the ones that change what a business can practically run. Published at the end of each month.
In preparationNothing published in this category yet.
Tell me what you run today and what you are trying to keep in-house. I will tell you plainly what is realistic on your hardware and what is not.
Start a conversation