personal data projects
Tiny Data Adventures take big, messy datasets (fossil records, museum catalogs, eruption reports, reef surveys) and make them accessible to everyone: easy to read for someone new to the data, and usable with a keyboard, a screen reader or without color.
Design and build, with Claude Code
4 projects, 1 in progress
2025 – 2026
I'm a product designer and a data enthusiast, not a data engineer. Claude Code gets me from a dataset to a working visualization fast; my eye and judgment decide what ships. Outside these solo projects, this is how I'd bring it to a team: running prototypes people can react to instead of static mocks, and small, well-scoped changes made directly in the codebase alongside engineers.
No Figma file, no design system up front. I designed these apps directly in code, because I wanted to see the data first. The system came out of the code afterwards, which is why so much of the work below is consistency and accessibility audits.
01 · The loop
Same loop, different lessons
Seven steps from messy data to something anyone can explore. Claude Code is in every step; what changes is what each of us brings. Pick a step, then a project in the tabs below it.
↺ 05–06 send fixes back to 04
Underneath every step:
02 · Context, not prompts
Memory and know-how
A CLAUDE.md in each repo is its memory; skills are routines that travel between projects.
Memory needs maintenance too. Tracing Origins' CLAUDE.md grew to about 44,000 words, full of reversals, and all of it loads every session. Treat it like code: keep it short, archive the history, delete what's superseded.
The three skills I distilled from these projects:
visual-consistency-qaAligns radius, color roles, spacing and type, so they can become tokens.
wcag-audit-learning"Learning" because I check every case by hand and talk each fix through before it's applied.
design-tokens-systemFixes contrast and sizes in one place, and holds each project's visual identity.
Each project's tokens page shows what is tokenized and what isn't yet.
/#tokens · Token Lab🌋 Volcano Explorer/styleguide🏛️ Tracing Origins/#tokensNext: a tokens page03 · Over time
What my commits say about how I work
Every commit from the four repos, one square each, placed by date. The marks are the moments my way of working changed; the band below shows what the commits were about, month by month. Before all this: a 2024–25 bootcamp, seismic-agent written with ChatGPT, and a coral project that stalled.
were about
Hover a square or a mark to read it.
Phases are tagged automatically from commit messages, so a few may be off.
04 · The projects
What each one taught me
Same loop, a different kind of mess each time. Pick a project to read its story.
05 · Takeaways
What I've learned working this way
Context over prompts
I talk to Claude like a coworker. The CLAUDE.md and the skills carry the precision.
Verify everything
Audits, data and every "it's fixed" get checked in the browser before I believe them.
Learn the fixes
I run the tests myself, so I understand each fix instead of just collecting it.