Infinite exercises, verified
LIVEA model drafts maths questions against the component library, a verifier throws out the junk, and a clean one renders. Forever.
system · Lumen8 (AI question generation with a verifier)
A model drafts a question against the component library. The verifier throws out anything without a single clean answer, then a good one renders. Repeat forever.
Nothing yet. Generate one.
Hit generate. Watch a question get drafted, checked, sometimes thrown out, and then rendered into a real interactive component. Hit it again for a different one. The point is that there is no fixed question bank: the supply is endless.
Why this is the interesting part
A maths platform lives or dies on having enough good practice. Writing thousands of questions by hand does not scale. So instead of a bank, Lumen8 gives a model two things: the catalogue of interactive components it can render into (planes, charts, fractions, and the rest) and the specific curriculum outcome a question should target. With that context, it drafts questions on demand.
The catch is obvious the first time you try it: a model left alone will happily produce a question with two correct answers, or a fraction that is already simplified, or a distance that is not a whole number and therefore miserable to mark. So a draft never reaches a student directly. A verifier checks every one: exactly one correct answer, difficulty inside the target band, and it has to actually render in a real component. Anything that fails is rejected and redrafted.
That is the whole trick. The model brings the variety, the verifier brings the trust, and between them you get something a fixed question bank never could: an endless supply of exercises that are still safe to put in front of a kid.