Project Write-Ups¶
The site maintainer's own accounts of real projects built with artificial intelligence (AI) assistance, written to show the method rather than the result: what was specified, what was checked, what broke, and what a colleague could reuse. They are software and data projects rather than clinical cases.
Where the AI Literacy Pathway teaches the foundations and the step-by-step guides walk through a task, these walk through a whole project after the fact, including the parts that did not work. Each one is written by the person who did the work, about their own work, and each is explicit about what it did not establish.
They are here because the useful skill is turning out not to be generating things with AI. It is specifying a project, deciding what the system is not allowed to do, verifying output you cannot fully read, noticing when the model is confidently wrong, and knowing when to stop. Those are teachable, and they are easier to show than to describe.
How to read these
Every claim in a worked example is meant to be traceable to something concrete: a file, a commit, a log, a published paper. Where a number was never measured, the text says so rather than estimating. Where something is unverified, it is labeled unverified. If you find a claim that does not meet that standard, the feedback form is the fastest way to tell the maintainer.
Color key: Guide
-
When a Check Stops Checking¶
What it is How this site is built, and the six weeks it displayed a product name that no longer existed
What it demonstrates Specification before code, a selection algorithm that turned out to be an essay, and five ways a check went on reporting success from behind its own blind spot
-
When Every Design Passes¶
What it is How this site was redesigned in four days by AI designers working in parallel, and how the scenery in its page headers is drawn in code
What it demonstrates Why a perfect test score cannot choose between designs, tasks held back from the designers, a second AI model as an adversarial reviewer, and tracing a real place from photographs
-
I Fixed Software I Cannot Read¶
What it is Building high dynamic range (HDR) screenshot capture into ShareX, a free screenshot tool, in a language the author does not write
What it demonstrates Deciding boundaries when you cannot review the code, refusing things on a maintainer's behalf, and an ending the author did not control
-
My Favorite Game Was Not a Game¶
What it is A media tracker and recommender for one family, built in three weeks without the author writing a line of it
What it demonstrates Testing software you cannot read, six ways a check passes against broken code, and a measurement that flattered its way into the profile
-
The Neanderthal Gene That Explained Nothing¶
What it is A widely reported genetics finding, tested against the author's own genotype
What it demonstrates Declining a flattering conclusion the data cannot support, refusals agreed in advance, and catching a confident error inside your own expertise
Content last reviewed September 3, 2026. Review dates are maintained as described on the About page.