The Process — a design-thinking case study for Sankofa Systems

Concept project — fictional brand. Sankofa Systems is a fictional generative art studio invented for this portfolio piece. There is no client, no gallery, and no sales; the site demonstrates capability. The algorithms, the code and the downloadable editions are real and yours to play with.

The build is part of a 26-site collection by Hannah Kwakye exploring one thesis: AI doesn't replace taste — it amplifies the designer who has it.

The client & the brief

Sankofa Systems is imagined as a one-room studio in Accra: two artists who write small programs that grow images, and who wanted a web presence that is the work rather than photographs of it. Their brief, as we framed it: a gallery where every visit hangs a different show; three distinct bodies of work drawn from Akan visual tradition; visitors able to keep what they find; and a commissions pipeline for generative identity systems. No stock imagery, no portfolio grid, no artspeak.

The problem

Generative art on the web usually fails in one of two ways. Either it is a screensaver — infinite, animated, unownable, gone when you close the tab — or it is screenshots of generative art, which concedes the whole point. The design problem was to make browser-grown work feel collected: singular, attributable, keepable. That reframed the project from "make pretty canvases" to "design the conditions under which an algorithm's output reads as an artwork" — which is a museology problem as much as an engineering one.

Creative direction — chosen, and the roads not taken

We chose the white cube: gallery-white walls, ink-only chrome, Instrument Serif museum labels, generous hanging space, plinth shadows — and all colour delegated to the works themselves through per-series algorithmic palettes. The chrome behaves like architecture; the art behaves like art. Four other directions were prototyped or argued through, and rejected:

Literal adinkra reproductionrejected

Rendering the actual sankofa and dwennimmen glyphs as decorative motifs. Rejected twice over: culturally, reproducing sacred symbols as website ornament is exactly the appropriation we wanted to avoid; artistically, tracing a fixed glyph leaves the algorithm nothing to say. We study the symbols' logic — the backward glance, the paired spiral, the strip-woven grid — and let the programs speak their own sentences. The studio section carries this position openly.

The dark dramatic galleryrejected

Near-black walls, spotlight gradients, gold everywhere. Seductive in a mock-up, but it fights the work: three algorithm families with their own palettes need neutral walls, and permanent drama flattens the one series (Night Return) that earns darkness by contrast. The art carries the colour, or the design does — not both.

Perpetual animationrejected

Works that grow forever in slow rAF loops. It demos beautifully and collects terribly — an endless piece has no final state to own, burns battery, and under reduced-motion preferences would need a completely separate still experience. We made stillness the design instead: a work renders once per seed, like a print pulled once from a plate. Regeneration is a deliberate visitor act, never ambience.

WebGL / shader renderingrejected

Fragment shaders could render richer texture, but the drawing model is wrong for this work: these are calligraphic, compositional pieces — ribbons, glyphs, woven blocks — which the 2D canvas API expresses in a few hundred readable lines. A shader pipeline would triple the payload and the guide page would teach GPU plumbing instead of drawing. The teaching artifact mattered more than the sheen.

Key decisions & trade-offs

Seeds over Math.random(). Determinism costs discipline (stable draw order, sub-streams for resolution-dependent texture) but buys the entire concept: editions are citable, reproducible, and honest. The label's seed is a promise — bring it back and we regrow your work exactly.
Two sliders, not twenty. Each algorithm exposes exactly two parameters (density-like and form-like). Real generative systems have dozens of knobs; visitors given dozens produce mud and feel lost. Two curated axes keep every reachable output inside the series' point of view — the artist's taste stays in the loop even when the visitor is driving.
The contact-sheet test. A generative series was only accepted when six seeds across the full parameter range could hang together as a wall. This forced the real work: frame-fit probing so nothing clips, a hushed ring around the focal glyph, opaque paper-mixed depth instead of muddy translucency. Judging single lucky renders is how generative galleries end up shipping their best-case and displaying their worst.
Export as museology. The download is not a screenshot — it re-renders at print resolution and typesets a museum label into the sheet (title, series, algorithm, seed, year, gallery). Trade-off: a second render costs a beat on old hardware; in exchange the kept object is archival rather than a souvenir.
Fictional studio, real contact. The commission form posts to Netlify Forms with a honeypot and an inline success state — no fake mailto, no dead endpoint — so the demonstrated pattern is exactly what a paying client would receive.

On the adinkra reference

Sankofa and dwennimmen are adinkra symbols of the Akan people of Ghana and Côte d'Ivoire; kente is the woven cloth of Asante and Ewe weavers. This project treats that vocabulary as inheritance to think with, not ornament to lift: the works never reproduce the symbols or the cloth, sources are named in the gallery itself, and the studio's stated policy is adaptation with respect — where a work drifts from it, it comes down. Designing from a living tradition means the tradition outranks the design.

Intended impact — as design goals

A visitor who has never heard the phrase "seeded PRNG" should leave able to explain why their edition is theirs — the label does the teaching.
The gallery should feel worth revisiting: every reload is a new show, and the show is never embarrassing at any slider position. That is a testable claim, and we tested it.
As a portfolio piece, it demonstrates a deliverable class few studios offer: identity systems that generate themselves, shipped as running code with export tooling — the commission pitch is the site you are standing in.

Designing a gallery that rehangs itself.