B. Koop / Product-minded engineer
Staff-level engineer with a designer’s eye for detail. I work across interaction design, design systems, and the engineering behind dependable products.
Selected work
Product design and creative experiments share a home here with the engineering that makes products dependable. Pick the kind of problem you want to explore.
Design & interaction
Routable · design system modernization
Contributed to modernizing Routable’s design system in a React product environment. The work sat at the intersection of reusable UI foundations and the needs of teams building real product experiences.
Personal prototype · visual exploration
An early 3D exploration of colored conversation paths with filters for messages and tool activity. The visual metaphor bends and kinks to invite a person to notice where their attention shifts. It is an exploratory interface, not a validated measure of focus.
Play Lab · browser prototypes
Slide-Tris and Spin Maze are compact playable experiments. Each explores how a small change to a rule or movement can turn a familiar interaction into a fresh puzzle.
Engineering & quality
Personal tool · AI-assisted reviews
Automated reviews grounded in reproduced behavior. Vette checks that a regression test catches the reported issue, then uses a blind secondary reviewer to look for missed cases. The review workflow includes TDD and mutation checks.
G2i · multi-repository work
Helped improve CI and pull-request workflows across multiple repositories. Reported pipeline time moved from about 60 minutes to about 20 minutes, giving teams faster feedback while they worked.
At Smartrise, I wrote a test plan and automated parts of the testing workflow. Monthly controller output grew from 60 to 300+ as the team expanded from 3 to 4 test techs, alongside additional wiring staff. At Chase, I helped nontechnical and technical colleagues learn to write runnable basic tests in a 30-minute session.
More about meField notes
A harmless collection of hidden prompt-injection examples for testing whether content is treated as content rather than instructions.
I built a small local dashboard to understand where my model tokens go, what they cost, and which kind of work consumes them.
We see them, and some people have made them, but what exactly goes into a good auto formatting input element
Have a good problem?