← Case Studies

Muse Group Design System

Design System Architect · Muse Group · 2024 – Present

Designed and built a cross-platform design system at Muse Group — the company behind Ultimate Guitar (30M MAU), MuseScore (100M yearly visitors), Audacity, and other world-leading music products serving 400M+ users globally. Covered token architecture, component primitives, documentation, governance, and measurable adoption across the entire product portfolio.

Problem

  • Inconsistent UI across Muse Group's product portfolio — Ultimate Guitar, MuseScore, Audacity, MuseHub, and others — spanning web and mobile platforms, with duplicated component implementations maintained by separate teams.
  • No shared token system — theming changes required manual updates across codebases, taking weeks per rollout.
  • Accessibility and visual quality varied between products, with no shared standards or automated enforcement.
  • New feature development was slowed by the need to rebuild common UI patterns from scratch for each product.

Constraints

  • Multiple products with different tech stacks and release cycles, collectively serving 400M+ users across web, iOS, Android, and desktop (Windows, macOS, Linux)
  • Existing codebases with significant legacy UI patterns that couldn't be replaced all at once
  • Teams with varying levels of design system familiarity and different workflow expectations
  • Need to support both web (React) and mobile platforms from a shared foundation
  • Scale demands: Ultimate Guitar alone handles 30M monthly active users; any component regression is immediately visible at massive scale

Role & Ownership

  • Designed the token architecture (DTCG specification): Figma Variables → JSON → CSS/TypeScript/iOS/Android pipeline
  • Defined component architecture: headless core primitives with product-specific composition layers
  • Built quality gates in CI: Storybook as source of truth, visual regression tests, Playwright interaction tests, axe accessibility checks, performance budgets
  • Established governance model: SemVer versioning, Changesets, RFC/ADR decision records, deprecation policy
  • Created the dev portal and knowledge base: MDX cookbook, component selection guidance, semantic search, usage analytics
  • Introduced AI agentic workflows across the DS lifecycle: automated component building and updating, Chromatic visual regression testing, unit test generation, component validation, and product UI generation from DS primitives — alongside a DS RAG assistant for engineers/designers and AI-powered PR linting

Options Considered

Incremental migration vs full rewrite

Chose incremental migration — wrapped existing components where possible and introduced new primitives alongside legacy patterns to avoid blocking product delivery.

Centralized ownership vs distributed contribution

Centralized architecture and governance with distributed contribution model — product teams can extend components through composition layers while core primitives remain centrally maintained.

Headless core vs fully styled components

Headless core with product-specific styling layers — allows each product to maintain visual identity while sharing interaction patterns, accessibility, and API contracts.

Outcomes

  • 95% token coverage across frontend; theming rollout reduced from weeks to hours
  • 70% reduction in visual and accessibility regressions by making tests CI release blockers
  • Unified React web components, reducing bundle size and maintenance overhead
  • Design system adoption increased 30% quarter-over-quarter through dev portal, starter kits, and training programs
  • Component reusability improved 40% through systematic architecture (during DS Engineer phase)
  • New feature development time reduced 30% (during DS Engineer phase)

Lessons Learned

  • A design system succeeds as a product, not just a library — adoption requires documentation, onboarding, and measurable feedback loops, not just good components.
  • Quality gates in CI are more effective than manual review for preventing regressions at scale.
  • Token architecture decisions compound — getting the DTCG pipeline right early saved significant rework across platforms later.