← All projects

Design system · AI

TCE Components 2.0

The TCE-MS institutional design system, built in Angular. I wrote the documentation that people and AI read before building a screen, and measured its effect in a blind test.

Organization
TCE-MS
Period
2024 · Ongoing
My role
AI-ready documentation, tokens and the Figma ↔ code bridge
Tools
FigmaAngularDesign tokensFigma MCPClaude CodeGitHub CopilotAgent Skills

Context

Analysts started prototyping screens with AI assistants. Without a clear reference, each AI made up its own design system: loose colors, components rebuilt by hand and library contracts used the wrong way, often with no visible error. The question was how to make AI actually use the court’s library and Foundations.

What I did

  • Each component documented from the source code, not from how it looks: a ready-to-copy example and pitfalls up front. Where the code has no answer, the text says so instead of guessing.
  • A single source: facts about the library live in one place, and every other instruction just points to it.
  • Blind A/B test before scaling: the same screen task, the same AI model (Claude Sonnet 4.6, through GitHub Copilot), three runs per group. Without the docs, 5 serious API errors per run; with them, none.
  • Priorities from real usage: counting components in the systems that already use the library, instead of documenting everything on a hunch.
  • Shipped with the package: whoever installs the library gets the docs, an instruction file and a prototyping skill in the open Agent Skills format, which works in Claude Code and GitHub Copilot.
  • Pilot with analysts, using a ready-made workbench to see the screen running.
  • Survey of library patterns and accessibility gaps, shared with the team.
  • Design tokens with naming and validation; Figma variables connected to code through MCP.
  • Documentation and Foundations released in the official library version.

© 2026 Fabiano Braz. All rights reserved.

Built with Angular · fabianobraz.com.br