About Me

I'm John Rioflorido, a design technologist working on AI evaluation and accessibility.

Safe AI requires accessible AI. People with disabilities are first-class citizens, and a system that hasn't been tested with them hasn't been fully tested. That conviction runs through my work: I measure whether AI systems do what people need, including the people a test usually leaves out, and I build the open-source tools that make that measurable.

I've spent more than a decade building accessibility tools that people use in production, and I'm now part of an AI innovation unit where I use those skills on AI tooling. My background spans design and development, cartography and accessibility, research and application, so much of my work happens where those fields meet.

Independent work

  • AI evaluation. A pre-registered eval of how AI coding agents receive written guidance, and what they skip when it doesn't arrive.
  • OKCA and klar. OKCA is a contrast algorithm that never approves a color pair WCAG 2 rejects. klar is the command-line tool that puts it in the hands of people and AI agents.
  • Candor. An open-source design system whose contrast minimums are written into its tokens and checked on every pull request.
  • Standards. Public comments to NIST, filed in a personal capacity, on AI agent security, AI evaluation design and AI documentation.

This blog is where I work out how accessibility and AI safety connect, on my own time and in a personal capacity. My ethical stance sets out the commitments behind that personal work. Like everything here, it's my own view; see the organizational disclaimer.

Contact Information