Case study
Evals and standards to improve an AI search assistant
A rule can be right most of the time and still fail the answer that matters. I named the failure modes, co-wrote the checks, and helped calibrate the judge that applies them.
Content design · Writing standards · AI evals
I write interface copy and the content standards behind it. Then I build the evals and skills that hold AI writing to the same bar.
Crafting and framing AI-generated content
Case study
A rule can be right most of the time and still fail the answer that matters. I named the failure modes, co-wrote the checks, and helped calibrate the judge that applies them.
Live tool
One standard should do two jobs: steer the model, then grade what it writes. Choose the standards and watch the judge score the difference.
Case study
Cohesion by default, difference where difference is true. A shared vocabulary for people, encoded as a skill AI agents can follow.
Case study
People trust an AI’s memory when they can see where each part came from. Interface copy and labels that separate what you said from what the AI inferred.
Writing and creating content systems
Case study
Every notification should tell a mover something worth acting on. I tested 43 variants with 55 users and turned what worked into rules for writers and AI.
Writing sample
Wrote first-ride content for a driverless car app, in-car screens, and in-car audio. Tested with riders; presented to stakeholders.
Writing sample
Kept what sales needed up front, saved the rest for after the letter, and gave the flow a plainer, warmer voice. (68 → 29 questions)
Writing sample
Developed guidelines for marketing, product, and support emails, built with those teams.