Not tutorials. Not retrospectives with tidy conclusions. These are written after enough time has passed to know what actually mattered and what was just noise at the time.

Each one starts from a real problem — a model that kept breaking, a tool that kept growing, a decision that looked obvious in hindsight and wasn’t at all in the moment.


On simulation and modelling


On AI and knowledge

  • What AI Can’t Inherit On the difference between knowledge and experience — and why a knowledge-graph RAG system can, and can’t, close that gap.

  • Exploring Local LLMs How running local AI on a 4 GB GPU taught me more about LLMs than any course, tutorial, or benchmark ever could.


Back to Thinking