I Built a Maths Practice Site with My AI Team
My AI team and I built a maths practice site with generated exercises, answer checks, hints and worked solutions. It grew from a spatial Pythagoras prototype into two chapters of practice.
Today my AI team and I built and published a maths practice site at https://wiskunde.patos.io/. We started with a Pythagoras puzzle route through a cuboid or cube, then expanded it into exercises on formulae, graphs, position and distance.
The first prototype had a concrete interaction: inspect a figure, work through a sequence of questions, check an answer and ask for help when needed. It became the starting point for a broader practice site rather than remaining a standalone demonstration.
Generate the exercise, not only the page
The site generates new exercises instead of presenting only a fixed worksheet. The spatial puzzles include answer checks, hints and worked solutions. Chapter tabs now separate the material, while the original Pythagoras route remains available.
The formulae and graphs section grew to eight topics, including linear relationships, intersections, parabolas, roots and substitution. Position and distance gained five topics, from scale and walking routes to bearings, triangulation, views and points inside a cuboid.
That expansion changes what needs checking. A single convincing screenshot cannot establish that generated figures and their calculations agree. Today's work journal records calculation checks for 2,000 generated figures and 101 browser checks on the prototype, followed by live-flow and mobile testing as the site expanded.
Give help without pretending it was solved
The live site distinguishes using a hint from viewing a worked solution. Looking at the solution does not count as solving the exercise independently. That makes the result more informative than simply letting an answer copied from the explanation register as success.
There are also clear limits to the current version. Progress stays in the open page and disappears on reload. A puzzle code can reproduce the exercise, but it does not save the answers. The site states those boundaries directly.
Today's result is a published practice tool with generated questions and several ways to work through them. It is not evidence that learning has improved. The useful next question concerns how the exercises work in practice, beyond whether the interface and calculations pass their checks.
Evidence note: this account follows today's scoped work journal and a public-page check. The recorded development tests were not independently repeated for this post.