Ways of Healing

From a reading list I couldn't finish to an atlas anyone can wander.

The idea began in a medical anthropology course I took in 2024. The material was deep and fascinating — but there was far too much reading for the time available, and I kept imagining an interactive website that could make learning it feel less like homework and more like visiting an exhibition.

Healing has never been only about seeing a doctor and taking medicine. It is bound up with family, faith, history, the communities we live in, and how we make sense of suffering. Yet when cultural healing traditions show up online, they tend to be either buried in academic papers or framed as something strange and exotic.

The core question became: what if learning how the world understands sickness and care could feel like wandering through a field archive, not studying for an exam?

My earlier interactive atlas, Where Dragons Dwell, placed 3rd in the 2026 Codex Creator Challenge and proved the map-based exhibition model. Ways of Healing takes that starting point into far more sensitive territory — illness, trauma, and care — with a completely different subject, audience, and content system.

One primary user wins the tie.

Early critique pushed me to stop listing possible audiences and actually choose one. When a design decision serves one group at the expense of another, this is who I protect:

1. 01

   Culturally curious general visitors

   Adults who find the topic meaningful or personally resonant — through immigration, multicultural families, or caregiving — and want a respectful, visual way in. They are not extracting facts; they are seeking recognition and understanding.

2. 02

   Educators and facilitators

   Instructors who can use the atlas as a discussion starter or pre-class exploration tool — a visual introduction rather than a textbook replacement.

3. 03

   Design learners and reviewers

   People interested in how complex academic content becomes an interactive system. They are an audience for my process, not the content — so they never drive the user experience.

Director and decision-maker in a human–AI studio.

I did not code this project line by line. I acted as director; AI (Claude and Codex) acted as researcher and builder. Every part of the site followed the same loop: I set the goal and constraints, the AI produced a draft or a working build, I reviewed it page by page, and the AI revised until I signed off. The AI never set direction on its own.

Three risk rules kept the style honest: not too decorative, not a treasure map, not cute — because the subject is illness, trauma, and poverty.

It started on paper.

I sketched three homepage directions — a flat world map with filters, an image-led gallery, and a rotating vector globe — then used AI to prototype each one before committing to the globe. The same sketchbook shaped the case pages, the Comparison Lab, and the Bookmark Workshop.

One catalog feeds every page.

The globe is the front door, but there are five ways in: cases, concepts, themes, comparison, and an AI guide. Underneath, a single source of truth (catalog.js) feeds shared scripts that render every page, so content stays consistent across the whole site.

Designed for reading, wandering, and taking something home.

Twelve wandering figures

Cut-paper "walkers" stroll along the bottom of each page — defined by what they carry, never by ethnicity. Click one and it stops to speak a short in-character line, like an NPC in a quiet game. Usability feedback later earned them an on/off toggle.

Herbalist (basket & sprig) · Lantern bearer (stigma stories) · Field scholar (map tube) · Water bearer (two buckets) · Caregiver (swaddled child) · Book carrier · Bowl bearer · Letter carrier · Medic · Elder · Parent & child · Dog walker

The hardest part wasn't code. It was truth.

AI drafted content from roughly forty-five medical-anthropology sources — and its output looked about 80% correct, which is the dangerous kind of correct. I ran three full audit rounds against the actual sources under four standards:

• No figure or description may appear that is not found in the source material.
• Everything must trace to the sources I provided, or to a reliable academic source cited instead.
• Cite title and author — never page numbers that can't be verified.
• Cited content must match its reference; no misattribution.

The audits removed invented statistics, corrected mistitled references, fixed numeric errors, and rebuilt the bibliography to 57 verified works. Where a figure couldn't be confirmed, I kept the statement general rather than invent detail. The same discipline applied to images: I approved or rejected all 51 illustrations one by one — an early HIV illustration was rejected because the AI had drawn a coronavirus.

Illustration rules were non-negotiable: no clear faces, no wounds or remains, no aestheticizing of suffering.

Two rounds: a survey, then a live test.

Round 1 — feedback survey (n=8, June 2026)

The written comments pointed one direction: readability. I enlarged type across the site, cut the handwritten typeface to reduce font noise, and demoted the "How to Explore" page out of the global nav to lower cognitive load.

Round 2 — moderated usability test (June 30, 2026)

A single-participant, think-aloud session with "Kevin," 40, an engineer and museum-goer — a fair proxy for the general-interest audience. One participant can't measure how common a problem is, but he can find real ones:

He didn't want to read instructions first — he wanted to begin exploring immediately. — field note from Kevin's session, and the best argument for demoting the how-to page

An exhibition that stays open.

Ways of Healing shipped as a complete, live atlas: 17 cases, 10 concepts, five entry paths, two scoped AI guides, and a souvenir stand — all in a visual language that treats illness and care with respect rather than decoration. The architecture assumes growth, so new cases can join the catalog without touching the design.

Two lessons I'm keeping. First, users won't necessarily accept a designer's favorite ideas — the walkers needed a toggle, and the evocative labels needed to become plain ones. Second, a designer paired with AI can multiply their output several times over — but only if the collaboration is transparent and the human owns the judgment, the sensitivity calls, and the accuracy standard.