← All work Alekhya.
Design System · Enterprise

A Design System for an aerospace company

Ant Design System overview

Turning 17 one-off button variants, 9 shades of the same "primary blue", and hours of redrawn modals into one shared language between designers and engineers.

Role
UI Designer, system owner
Responsibilities
UI audit Token architecture Component design Documentation Dev handoff Adoption support
Project
Aerospace client, three parallel products
Timeline
3 months, 2023
Tools
Figma, InVision, Storybook
Context

Three squads, three products, three different buttons.

A leading aerospace company shipped three parallel enterprise products built by different squads. Every new sprint, designers were rebuilding the same buttons, modals, and tables from scratch, each with a slightly different hex, a slightly different radius, a slightly different hover state.

I was brought in to turn that into one shared language between design and engineering, built on the ANT-V 14.3 foundation. The brief was deceptively simple: make the Figma file and the shipped UI the same thing.

Style guide overview
The Problem

The audit was the finding.

Before any design, I ran a quick audit: interviews with 4 designers, 3 frontend engineers, and a visual inventory across the three live products. The numbers were damning on their own:

  • 17 different button variants across the three products, 4 of them used only once.
  • 9 shades of the same "primary blue", no two were an exact match.
  • ~40% of a typical sprint spent re-making components that already existed somewhere in the org.

The system had to replace that waste, which meant it had to be easier to use than ignoring.

UI inventory audit

UI inventory, a sample of the duplicate component variants surfaced during the audit.

Research

Designers and engineers wanted the same thing, in different words.

The interview quotes told a sharper story than the numbers:

"I build a modal, ship it, and three weeks later the other team ships a different modal. Neither of us knew the other was doing it."Frontend engineer, Product B
"I don't want a style guide PDF. I want something I can drop in Figma and know it's correct."Senior designer, Product A
"Half the specs I get from design are wrong because they're copied from an old file. If the file was the source of truth, that would just go away."Tech lead, Product C

"Drop in and know it's correct" became the usability bar. Zero docs needed for the top 10 components.

Design Decisions

Five decisions that made the system usable.

1.

Atomic structure, atoms to organisms.

Not for theory, for review speed. When a developer contested a spacing value, we could review the token once, not re-check every screen that used it. Atomic structure made the system auditable in a way a flat library wouldn't have been.

2.

A hard cap of 10 shades per hue.

The audit found 23 near-identical greys. Two testing sessions showed designers couldn't reliably pick between adjacent shades anyway. Capping at 10 forced intentionality and gave developers a finite token list to implement.

3.

All five states documented upfront.

Default, hover, focus, filled, error, disabled. Missing states were the single biggest source of design-vs-shipped discrepancy. Making states mandatory on first spec cut the "can you add disabled?" back-and-forth to near zero.

4.

Docs sit beside the component, not in a separate wiki.

Three interviewees said they never read the old wiki. Inline, co-located docs get read because they're in the way of work, not a side quest. Each component in Figma carried its own usage, do's, and don'ts right next to it.

5.

Incremental migration, not clean-slate.

A clean-slate design system was faster to build but would have stranded live products on the old components. Incremental migration (ship new component, mark its old siblings deprecated, let teams migrate on their next sprint) was slower but meant nothing broke in production while the system rolled in.

Final color palette part 1 Final color palette part 2

Final colour palette, consolidated from 9 near-identical blues down to one primary scale with 10 defined steps.

Documentation example Documentation example
Documentation example Documentation example

Component documentation, co-located with each component, so specs and rules travel with the asset.

What Didn't Work

Two things I tried that the team rejected.

  • A "component request form." I set up a Notion form for teams to propose new components. Two requests in four weeks. Replaced with a 30-minute weekly open hour, 11 requests in the first fortnight. Live conversation beat a form.
  • Enforcing the system via CI lint rules. We tried blocking merges that used off-system colours. Developers hated it, it blocked urgent hotfixes. Switched to warnings in PR comments; adoption climbed because it stopped feeling like a gate.
Reflection

The system shipped, and the team started composing screens instead of building buttons.

Measured 6 weeks after rollout, across the three product squads:

  • Design-to-dev handoff time dropped ~42% on component-heavy screens; average spec turnaround went from 3.1 days to 1.8.
  • Button variants in use fell from 17 to 6, the 6 that matched a real behavioural distinction.
  • 88% of new screens shipped by week 8 used only system components (baseline was 0%).
  • Qualitative shift: two designers moved from "I build a button" to "I compose a screen"; it freed them for the harder interaction problems the system couldn't solve for them.
  • Adopted as the default for all new product work in the squad, and extended to a 4th product in Q2.

The lesson I carry out of this project: a design system is a social tool first, a Figma file second. The hardest part wasn't building components; it was building the trust and the habit for people to actually pick them up.

Limitation: I didn't run quantitative accessibility audits (contrast, ARIA, keyboard) on every component before handover, the timeline didn't allow it, and a11y became a follow-up tracked by the squad. In retrospect, carving even a day for a programmatic a11y pass would have been worth the trade.