The meeting is a first consultation. I build prototypes, POCs and production-quality code, so whether you have an idea you want to see visualized or something ready to deploy, I can help. I also consult with teams on AI and L&D.

I created the spec and the phased build plan by researching the problem, collecting benchmarks, sketching in Figma and pulling context out of a second brain I built. I crafted and honed it with reference documents, long planning and my own reading of what the model gave back, before any of it became code. An agent wrote the code. Deciding what was wrong is the last step, not the job.

The spec lives at the root of the repo as CLAUDE.md. Tokens down to the hex, component contracts, a phase list where nothing starts until the previous thing passes its check. It took longer to write than the first few pages took to build, which sounds like bad maths until you've watched an agent confidently invent a component library nobody asked for. The specificity is the job. Vague instructions still get you vague software, just faster than before.

The method is the one I teach: stream, structure, guide, build. Dump everything, let a model turn the mess into a brief, argue with the brief until it's small enough to be true, then build one testable thing at a time. So the four-phase workshop on the enablement page isn't a retrofit. It's what produced this site. The palette you're looking at is the second one, incidentally. The first was monochrome and sharp-cornered and I liked it right up until I put it next to something warm.