Case study
A design system for 100,000+ users
100+ components, 30% less development time, one source of truth.
- Role
- Software Engineer II, SYDLE
- Period
- Jun 2022 — Mar 2023
- Stack
- StencilJS
- Web Components
- TypeScript
- Design Tokens
- SSR
TL;DR
- Built a design system of 100+ components in StencilJS and Web Components, so one implementation served products on different stacks.
- Cut application development time by 30%, and underpinned a BPMN marketplace on a platform with 100,000+ users.
- Refactored the blog to server-side rendering: 80% faster, and 40% more sales conversions.
Key figures
- Components shipped
- 100+
- Less development time
- 30%
The problem
The company's products were built on different stacks, and each of them was rebuilding the same inputs, tables and dialogs. Every copy behaved slightly differently, every fix had to be made several times, and a product's look depended on which team had built it.
A component library for one framework would have helped one product. The problem was that there was more than one framework.
Constraints
- A small team for a company-wide system. Six people: a designer, a tech lead, and four developers including me.
- Products could not stop for it. Adoption started with new products only; older projects migrated gradually afterwards.
Decisions and trade-offs
Web ComponentsoverA framework-specific library
A custom element is understood by every framework and by none in particular, so one implementation can serve all of them. The design system was framework-agnostic by construction rather than by wrapper: a product adopted it without a rewrite, whatever it was built on.
StencilJSoverHand-written custom elements
Stencil compiles to standard Web Components, so the output carries no framework dependency, while authoring stays in TypeScript and JSX with lazy-loading built in. At more than a hundred components, that authoring experience is what kept the library maintainable.
Design tokensoverStyles inside each component
Colours, spacing and type were defined once as tokens and consumed by every component, so a visual decision is made in one place and lands everywhere.
Using it myself
I was also one of the system's consumers. I built the SYDLE marketplace in Angular on top of the components, which is the framework-agnostic decision tested from the other side: a product team picking up custom elements and shipping with them.
Beyond the components
The same role covered two other pieces of work worth naming. I refactored the company blog to server-side rendering, which made it 80% faster and lifted sales conversions by 40%. And I migrated legacy JavaScript codebases to TypeScript incrementally, without freezing feature work while it happened.
- Blog load time100 → 20-80%
- Sales conversions100 → 140+40%
Results
- 100+ components shipped, serving products on different stacks from one implementation.
- 30% less application development time, measured by comparing the time spent on development tickets before and after adoption.
- A BPMN marketplace built on the system, used by a platform with more than 100,000 users.