Skip to the case study

Katalyst design system

Koddi's bespoke in-house design system.

Role
Initiated the effort, got buy-in from engineering and product directors, rallied the design team to contribute to it and adopt it
Team
Timeline
Jul 2025 to now
Impact
50% less designer ramp-up and design time

The problem

Bringing Koddi into the 21st century.

When I first joined Koddi, all of our products looked like Frankensteins of each other. The design system was a Chakra UI library that had not been updated in three years, so designers redrew the same parts for every project and engineers rebuilt them, slightly differently each time.

Engineers were overriding the button component or building their own. A designer looking for a component could not tell whether it existed. The question became: how do designers, engineers, and AI tools share one trustworthy answer for how Koddi looks and behaves, so new work is consistent by default instead of cleaned up afterward?

How it came together

  1. A weekly meeting, one component at a time

    In July 2025 designers and front-end engineers started meeting every week to work through components: shadows, KPI cards, switches, tooltips, date pickers, charts. We wrote a handoff checklist with it: labeled flows, annotations in Figma, and every engineer added to the design system file.

  2. Break things fast, fix faster

    The beta shipped in November 2025. From then on, breaking changes needed a ticket and a review. I asked that everything new be built in Katalyst so nothing would need redoing. The rule we agreed on: add variants as they're needed, but be careful once something is in use.

  3. Rules with reasons

    Components need behavior, not just looks. A modal is for focus when the page behind it is not needed; a drawer is for settings you check against the page. When a client design needed a custom colored card, we learned the system could not take outside colors without hacks, and decided to limit theming rather than bend every component.

  4. Designers become contributors

    In 2026 designers started shipping to the library itself. I opened a pull request for a search modal that became the omnibar component. A designer cloned the repo and launched Storybook in a meeting with an engineer guiding her. When engineers kept overriding a plain text button, it became a new variant instead of another one-off.

  5. The rulebook AI follows

    Katalyst is now wired into Figma and Claude Code, so a prototype looks and feels like Koddi from the first screen. Now anyone at Koddi, technical or not, can ship code that looks like our platform.

Results

50%

less designer ramp-up and design time, with a shared, documented library instead of one-off recreations

43

releases by October 2026, a year after the beta

20%

increase in engineering efficiency due to fewer one-off creations and UI fixes

Looking back

What I learned

A design system is three things, and they drift: the Figma file, the coded library, and the rules each live somewhere different. Our working rule became "if it's not in Storybook, it isn't a component." Components are the easy part; agreement between design and engineering is the work.

What I'd do next

Track adoption from day one, so I can say what share of screens use Katalyst, not just the version number. Keep Figma, Storybook, and Code Connect in sync as one task, not three. And decide the theming policy before the first client-colored design arrives.