GLDB: Bringing Consistency and Accessibility to a Design System
A shared foundation for a Singapore digital bank, with 36 accessible components across 300+ screens.

The challenge: one system, three problems
Linklogis was expanding overseas, but its design system, Link UI, existed only in Chinese. Instead of translating it, we built Link UI-International as a new English-first system on Element Plus, the Vue 3 component library.
- An inherited style guide that was drifting. It used 22 text colours, 28 corner radii and 11 font families, with only 2 reusable components. The primary button reached just 1.9 to 2.8:1 contrast against the 4.5:1 WCAG AA needs.
- A Chinese-first system. English labels in GLDB’s requirements ran about twice as wide as their Chinese versions. Fonts were tuned for Chinese, and interaction habits followed mainland China conventions.
- Many stakeholders with different tastes. Founding partners and product managers each had their own views, and without a shared standard, reviews turned into debates about preference.

Audit first, then build the rules
I audited the inherited file (41 boards) and mapped the product requirements, so the system was built around real screens instead of an abstract component list. With just me and one front-end developer on a tight timeline, we built on Element Plus: tokens restyled it to match GLDB, and a component joined the library only when two business modules needed it. Then I built the system in three layers:
- Tokens. Colour, spacing, radius and type were kept as Figma styles, changing the visual style as little as possible while closing accessibility gaps.
- Components. Atomic design, from atoms to pages, ending in 36 core components.
- Handoff. Each component went through six steps with the front end: tokens, specs, naming, edge cases, design quality review, and documentation and versions in Modao, so the Vue 3 build matched the design.


Make English the default
Each localization problem had a rule inside the system.
- Room for English. English runs wider than Chinese, so each component was designed with its longest English label first and left room to grow. Long text first wraps to a second line. If it still does not fit, it ends with “…” and the full text appears on hover.
- Type. Inter for English and Noto Sans SC for Chinese. Type was set up for English first, with size, weight and line height saved as text styles, and Chinese checked against the same styles.
- Density. Spacing and tables were set for English-length labels, keeping key columns visible and moving secondary actions into menus.
- Tokens named by meaning. Colour, type and spacing are named for what they do, such as “error” or “success”, and tied to no market or language.
- One vocabulary. The product managers’ spec document fixed the English terms for business concepts, such as buyer and core enterprise, and components and screens used those exact words.
- Accessibility from the start. Contrast, focus states and text labels were built into components, so meaning never depends on colour alone.
Let real screens decide
I shared work regularly and showed changes inside real product screens. A colour or style preference became a question about a shared rule, and reviews moved from debating individual choices to confirming a direction.


Outcome
- 36 core components
- 300+ high-fidelity screens across two portals
- MVP core features delivered in Nov 2021
- About 30% less front-end development time, by the developers’ own estimate
- I named owners and documented how to contribute before handing off, and Link UI-International went on to support Linklogis’s later overseas products.
Reflection
A design system needs agreement as much as it needs components, so setting shared rules before scaling page design was the core of the lead role. Going global added a second lesson: localization lives inside tokens and components, so it has to be decided there once.
Next time, I would follow what mature design system teams do:
- Agree on governance first. Before building, settle with the partners and product managers who own the system, who approves changes, and how a request becomes a component. Taste debates, then have a process to land in.
- Measure from day one. Our 30% savings were an estimate from the developers. Tracking component usage and build time from the start would have turned it into a measured result the business could see.
- Listen to the people using it. After each release, ask designers and engineers what helped and what slowed them down, and feed their answers into the roadmap.

