1
Design before code
A written design with numbered requirements, and a rough sketch of the screen agreed before anything is built. Nothing had to be rearranged afterwards.
Work · Case studies · Hire me
I design it.
AI tools build it.
Tests keep us honest.
I am a designer of apps and systems who builds with AI coding tools. I decide what gets made and what counts as correct; the tools write the code; everything is measured, and what went wrong is written down next to what went right.
Case studies
Two builds, with the numbers
Each was written while the work happened, not afterwards.
Desktop app · on sale
SnapShelf
A paid Windows and Linux app, from a design document to a product on sale in two days: more than 400 automatic tests, installers proven on fresh machines, its own license server, and a record of every mistake on the way.
Read the case study
AI assistant · evaluation
A handbook Q&A bot
A bot that answers from documents, cites the page, and says so when the answer is not there. Scored perfectly on the questions it was tuned on, and 4 of 6 on questions it had never seen. That gap is the finding.
Read the case study
Method
How I work
The same five habits on every project.
2
Small steps, each one checked
Work is cut into milestones with an exit test. Each is built, tested on Windows and Linux automatically, and reviewed before the next begins.
3
Check the research
AI research is treated as a claim, not a fact. A second pass found 7 errors in 33 researched items in one design before any code depended on them.
4
Test the real thing
Stand-ins are not enough. The first real call to a deployed server failed twice in ways no test had shown, in a sandbox, before any customer existed.
5
Write down what went wrong
Every project keeps a dated log and a case study, including the mistakes: mine, the tools', and the research's. You get the same record for your project.
What I can build for you.
Small, well-tested software with a clear job: a desktop tool for a repetitive task, an assistant that answers from your own documents and admits what it does not know, or a product page with payments and license keys behind it. You get the working thing, its tests, and a plain-language record of how it was built and what it can and cannot do.
What to expect.
I am early in this work, and these two projects are the evidence I have. I will tell you when I am unsure, when a number is small, and when something has only been tested against a stand-in. If your project is not a good fit for how I work, I will say so before we start.