AI design engineer. Prev @bayer @qiagen @philips @marsglobal - https://teapawt.org/t.co/drpSbnj48y
Building in public
Joined July 2020
- Tweets4.7K
- Following1K
- Followers5K
- Likes1.3K
A usability test isn't a usability engineering file.
For a medical device, a few good usability sessions aren't enough. You have to identify the tasks that could harm a patient or a user, and document the design control for each one.
Start with the tasks where a use error could harm someone. The FDA calls these critical tasks. A missed alarm or a wrong dose setting can harm a patient, so you design and test for those first.
If you're not sure your usability work would hold up as a usability engineering file, I can map your gaps against IEC 62366-1 and FDA human factors guidance in my usability engineering readiness review.
awolesigideon.com/work-with-…
The three-screen test I run to see if a coding agent can build from your design system.
I ask a coding agent to build three real screens using only your design system. Then I check each screen and trace every mistake back to its cause.
I also check your Code Connect coverage and your token naming, and you get a roadmap split into now, next and later, so you know the order to work in.
The AI-ready design system audit takes three weeks and starts from €4,500.
awolesigideon.com/work-with-…
Why safety demands retesting new flows in usability engineering.
When I wrote the test scenarios for a pulse oximetry project, I separated the SpO2 tasks carried over from an older product from the new safety behaviours.
And if you change a carried-over task at all, even a label or the order of the steps, treat it as new and retest it.
Need help planning your retests? My two-hour triage is €1,000, and it's credited if you book the full usability engineering readiness review within 14 days.
awolesigideon.com/work-with-…
Make your tokens human readable. Here's why.
Right now, a token called blue-500 only describes the colour. A designer reading the name still has to work out if it belongs on buttons or links, and your coding agent has the same problem.
And when you ask a coding agent for an error message, text-error is an easy match. With red-600, the agent may use a red that's meant for something else, like a delete button.
If you want me to check your tokens before a coding agent builds from them, it's part of my AI-ready design system audit.
awolesigideon.com/work-with-…
