Systems that make AI useful to real people.
Not slideware. The things below are running: a book you can hold, a funnel you can walk, courses that refuse to teach what they can't ground, and an agentic line that ships behind its own gates.
The throughline
AI enablement is a systems job, not a demo job.
Anyone can show a model doing something impressive once. The work is making it repeatable, grounded, gated, and maintained — so the impressive thing keeps happening, for real users, without a human babysitting it. Each system below is one piece of that discipline.
Six systems · what each one proves
A callable AI-publishing pipeline
“A Mind of Our Making” — a print-ready POD book on the history of AI — is produced by a deterministic Typst pipeline with blind critic panels, an arc grader, and per-figure quality gates. The pipeline is the artifact: research → storyboard → draft → copyedit → produce → grade → sign-off, callable as one command.
Proves · Turning a craft into a repeatable, gated system.
A federated knowledge layer with a source contract
A 55k-paper research corpus and a curated wellness graph, federated under one GRADE evidence bar. Courses are built by a harness that must cite graded sources or halt — a lesson that makes a factual claim ships its citations or it does not ship. Capability atlases map exactly what the corpus can and cannot ground.
Proves · Grounding AI output in evidence, by construction — not by hope.
An agentic delivery line, gated by a judge ladder
Coding agents take an issue to a pull request autonomously, behind an event-driven bus and a ladder of LLM judges (citation, visual canon, diagram beauty, reading flow). Every change lands behind required checks; nothing merges red. Built to be observable and reversible.
Proves · Letting agents move fast because the gates catch what they break.
A funnel from a printed page to a measured level
A QR code in the printed book resolves to an AI-fluency assessment, then routes the reader into free foundations and on into the course ladder — one continuous, instrumented path from atoms (the book) to outcomes (a level you can act on).
Proves · Designing the whole learner journey as one system, end to end.
Live, multiplayer learning
A presentation harness runs cohort sessions in a Jackbox style — teams, rounds, a host depth-dial — on a real-time service. The same content that teaches asynchronously also runs a room of people live.
Proves · Meeting learners where engagement actually happens.
Self-watching controls
A scheduled agent regrades every deployed course weekly, regenerates the capability atlases, and routes any regression into a work queue — so quality drift surfaces automatically instead of going dark. The systems above are not just built; they are maintained.
Proves · Building the control loop, not just the feature.
Start where it starts
The fastest way to understand the work is to walk the funnel yourself.
Measure your fluency in five minutes, then step into the foundations the same way a reader of the book does.