Vela is an AI-first design system

Vela

Overview

Vela is an AI-first design system platform. The same 63 components adapt through tokens — never forks — across shape, surface, brand, and eventually industry. Built to be read by designers, developers, and coding agents alike.

Main Problem: As products grow, design systems often become fragmented — with duplicated components, inconsistent patterns, and unnecessary variations.

I built a flexible, scalable design system that provides a single source of truth while adapting to different products, brands, and visual styles.

View in Figma

This resulted in:

  • A token-based system — flexible and easy to customize
  • An accessible system — built with WCAG 2.2 AA in mind
  • A developer-friendly system — easy to implement and maintain
  • A scalable foundation — consistent across different products and use cases

Role

User experience design
User interface design

Tools

Figma, Claude, ChatGPT

Team

  • 1 Product Designer (me)

Architecture

Every decision lives in one of four layers
A single component keeps consistent anatomy and behavior — shape, surface, and motion reconfigure entirely through tokens, so no one ever needs to duplicate a component just to reskin it. The status badges below are real progress, not a roadmap wishlist.

Why layers, not a flat list: I couldn’t let anyone start on Style Systems until Foundations and the Component Library were both provably solid — otherwise the original forking problem would have just moved up a level.

01 · FOUNDATION

Complete

Tokens, color, type, spacing, radius, elevation, motion, icons, accessibility.

02 · COMPONENT LIBRARY

Complete

Actions, Forms, Navigation, Feedback, Containers & Overlays, Content & Data.

03 · STYLE SYSTEMS

Validated · 3 components

Shape (Sharp / Rounded / Pill / Squircle / Brand) and Surface (Flat / Gradient / Glass) — restyled through tokens.

04 · INDUSTRY SYSTEMS

Banking pilot · 3 components

Banking, Healthcare, Enterprise SaaS, and more — inheriting from the core, never forking it.

Token resolution:Buttonsemantic/action/primaryprimitive/color/brand/600#6C4CF2

Foundations

160+ variables across 5 collections

Everything else in the system is built on these — components never hardcode a raw value.

132
Primitive
25
Semantic
1
Brand
2
Shape
6
Surface
Color — primitive scales feed semantic tokens

 

NEUTRAL 0 → 1000
 
 
 
 
 
 
 
 
 
 
 
BRAND 50 → 900 — brand/600 = #6C4CF2
 
 
 
 
 
 
 
 
 
 
DANGER 100 / 500 / 700
 
 
 
SUCCESS 100 / 500 / 700
 
 
 
WARNING 100 / 500 / 700
 
 
 
INFO 100 / 500 / 700
 
 
 

Components reference semantic tokens only — background/primary, text/secondary, action/primary, feedback/success, focus-ring — never a raw hex.

Typography — Inter on a type scale

xs 12sm 14md 16lg 18xl 202xl 243xl 304xl 365xl 48
STYLETOKENUSED FOR
Display5xl / BoldMarketing hero, cover
Heading / lg3xl / BoldPage titles
Heading / md2xl / BoldSection headings
Heading / smxl / SemiboldCard / subsection
Bodymd / RegularDefault paragraph
Captionsm / RegularMeta text, timestamps
Labelsm / MediumForm labels, buttons
Spacing & Grid — A single scale, one grid

A single spacing scale drives padding and gaps — plus the responsive grid it sits on top of.

A single spacing scale drives padding and gaps — plus the responsive grid it sits on top of.

4
8
12
16
24
32
48
64
TIERMIN-WCOLSGUTTERMARGIN
Mobile0px416px16px
Tablet768px824px32px
Desktop1024px1224px64px

Shape & Surface

One shape token, five personalities
The same component keeps its anatomy — only the shape token changes.

Sharp
0px — enterprise, editorial
Rounded
4–12px — SaaS, dashboards
Pill
Full — consumer, social
Squircle
8–16px +smoothing — mobile-first
Brand
Custom vector — identity moments

Three surface treatments, one hierarchy

Flat

surface/default — enterprise, dense interfaces, accessibility-focused. The safe default.

Gradient

brand/500 → brand/700 — hero sections, promo. Structured stops only.

Glass

backdrop-filter: blur(20px) — optional layer, needs a Flat fallback.

 

Never stack Gradient or Glass on dense, data-heavy, or form-heavy UI — tables and inputs stay Flat.

Why this stayed a token, not a component fork: an Enterprise SaaS venture and a consumer venture can now share the same Button, Card, and Input — the only thing that changes between them is which shape and surface token they resolve to.

Elevation & Motion

Depth and movement, both token-bound

0
Cards, rows
1
Hover-lift
2
Popovers
3
Sticky headers
4
Dialogs
5
Modals
DURATION
120ms200ms320ms480ms
fast · base · slow · slower
PERSONALITY MAPPING
Subtle → fast–base, standard/decelerate — enterprise, dense UI
Expressive → slow–slower, emphasized — consumer, wellness, promo

Iconography

Phosphor, mapped to personality
39 real Phosphor icons, token-bound. Weight maps to shape/surface personality; size and color always come from tokens.

SHAPE + SURFACEWEIGHT
Sharp + FlatRegular / Bold
Rounded + FlatRegular
Pill + GradientFill / Duotone
Squircle + GlassLight / Duotone

Component Library

39 patterns, six categories, every state specced
Grouped by category rather than one page per component — each ships every interaction state (Default/Hover/Pressed/Focused/Disabled) across Small/Medium/Large, all reading the same shape and elevation lookup tables above.

05
Actions
  • Button (Primary/Secondary/Danger)
  • Icon Button
  • Link
  • Button Group
  • Split Button
15
Forms
  • Text Input · Textarea
  • Checkbox · Radio · Switch
  • Select · Combobox · Date Picker
  • Slider · File Upload
  • Credit Card · Phone Number fields
09
Navigation
  • Tabs · Breadcrumbs
  • Pagination · Stepper
  • Menu · Dropdown · Command Menu
  • Sidebar · Top Navigation
07
Feedback
  • Badge · Alert · Toast
  • Progress Bar · Spinner
  • Tooltip · Skeleton
05
Containers & Overlays
  • Card · Accordion
  • Dialog · Drawer · Popover
10
Content & Data
  • Avatar · Divider · Tag
  • List Item · Empty State
  • Table · Tree View · Timeline
  • Calendar · Data Grid · Charts

Accessibility

WCAG 2.2 AA, built in
Every ratio below is computed from the actual tokens in this file.

PAIRINGLIGHTDARK
text/primary on bg/primary17.75:1 AA17.75:1 AA
text/secondary on bg/primary6.4:1 AA10.61:1 AA
text/disabled on bg/primary2.52:1 fails2.77:1 fails
icon/default on bg/primary9.45:1 AA13.46:1 AA
action/danger on bg/primary3.91:1 UI only4.54:1 AA
focus-ring on bg/primary5.31:1 AA4.8:1 AA
24×24
WCAG AA minimum
44×44
This system’s default

action/danger fails AA for normal body text in Light mode — pair error text with an icon.

Results & Reflection

Turn heads, not screws
One foundation, 39 patterns across six categories, and a token model built to be read by people and coding agents alike — replacing what used to be a separate library per visual style.

39
Patterns across six categories
5
Foundational token collections
160+
Tokens — zero hardcoded values
0.2.0
Current beta — still evolving
What I’d repeat
  • Foundations before components

    Every component page reads from the token pages built earlier — Shape’s lookup table, Elevation’s paired z-index, Motion’s personality tiers. Building Actions first would have meant re-deriving all of it mid-stream.Writing every decision as a resolvable token made the file legible to coding agents, not just designers

  • Reject native features that don’t ship

    Figma’s native Glass effect looked better in the editor but had no Dev Mode export — using Background Blur + translucent fill instead meant the spec was never aspirational.

  • Compute, don’t estimate

    Running real contrast math against the actual token values (not eyeballed) surfaced the one pairing that would have shipped as an accessibility bug. 

Next