Product designer for AI.
Systems thinker
by practice.

Frontier Cartography, the house

Brought the instruments under one roof at this address, each on its own subdomain, with a front page that states what each thing is and what state it is in rather than what I hope it becomes. The brand system — paper, ink, one brick accent — holds the work together. This revision makes my product-design role and specific contributions easier to find.

A portfolio is an instrument too. If it cannot expose its own state, nothing in it should be trusted to.

AI Council

A local decision workspace developed with AI assistance. I specified reusable roles, visible disagreement, decision history and human approval points, then questioned a partial run that needed repair. Codex contributed substantial technical design and implementation assistance. Independent specialist briefs feed a chair’s recommendation; the principal records a separate decision.

The verified synthetic example preserves differing positions and decision revisions. It tests the record mechanics, not whether a council improves human decisions.

ThoughtGlass

ThoughtGlass explores a reflective interface for a single inquiry. Its design commitments include a consentful blank start, provisional machine interpretation, human correction, and retained history. The linked worked example demonstrates a supplied correction, a structural preview, a graph redraw and a new question. It opens with an existing inquiry and does not demonstrate interpretation of arbitrary corrections; the blank-start tablet draft is a separate experiment.

Disagreement is not noise to remove. Designed properly, it is the next state of the system.

Explore the contract replay →
A guided replay of an exchange between Krystel, Claude, and Astra: two readings, human confirmation, and the questions that changed.

StarGlass

A depth-psychological astrology interpreter in the Greene tradition: deterministic ephemeris math underneath, generative interpretation on top, the chart machinery kept out of the prose and into the margins. Built twice by design — as a plain-file skill that travels between AI environments, and as a deployed web experience with its own ephemeris service. When I noticed the voice drifting after a week of readings, I built a rerunnable drift harness rather than editing by feel. When an outside review found I had exposed an unauthenticated API, I closed it and put a watcher on it.

One concept can live in many bodies — as long as the protocol and the product are not allowed to quietly become different things.

“I Spent 3½ Years Teaching a Machine My Language”

The essay that names the practice: private language, context packs, and what a model has to be told before meaning arrives intact — written from three and a half years of notes on what survived the trip and what didn’t.

The vocabulary was never the hard part. The hard part was noticing which words I thought I had already explained.

The AI Design Atlas

The Atlas turns a periodic-table metaphor into a language for learning with creative AI. Elements name a tool’s role, strength, and failure mode; bonds describe connections, including where a relationship is still unmeasured. A combination can become a structured Word one-pager, branded PowerPoint deck, infographic, meme, Mermaid diagram, or self-teaching protocol. The protocol offers a first attempt and a place to record how far you got and what you observed. The Watchlist keeps emerging candidates and retired tools visible.

The purpose is to understand more deeply, remix deliberately, and build something new — with a record of what you tried and what you still need to learn.

Signal and Soil

Filed as an Arizona nonprofit corporation: the non-profit counterpart to this practice, aimed at rural Yavapai County — skills, infrastructure, and the kind of community project where people make things and teach each other. The for-profit and the non-profit are deliberately separate bodies with one intent.

Meaning also has to survive transfer between an idea and a legal structure. The paperwork is part of the design.

UX RAVE

A Figma Make prototype for turning research context into structured artifacts. The prompts specify a format and prohibit invented metrics, participant counts, and quotes; the app renders the response into a readout. Those instructions express the intended integrity rules. The generated content still needs review. It grew out of an earlier app where the model returned structured JSON and the interface did the rendering — a generate-payload / app-renders pattern I keep reusing.

A format contract and a truth check are different jobs.

Teaching a machine my language

Kept a notebook while working with language models: which of my words a model understood, which it silently replaced with its own, and what I had to say to close the gap. The notebook became the context packs; the context packs became the practice.

A private language is a design artifact. Left undocumented, it is a liability with a nice vocabulary.

Twenty years across the disciplines

Graphic design, web, marketing, print, product, service design, and eventually enterprise architecture — in that rough order, each one a different receiving side. Before any of it, a corporate sales floor at a large for-profit university, where I was formally trained in persuasion and learned, from the inside, what happens to a script when the person delivering it stops believing it.

I learned to read the power structure under any visible arrangement — a sentence, a screen, an org chart — because I had once been asked to hide one.