Gaplight

NMT.academy is my own platform for Ukraine’s national exam. It is where the engine was built and, more to the point, where it is measured on real students every week.

What that measurement says right now (October 2026): across 49 students and 114,000 answers, the median student makes 43% fewer mistakes on the same material. Two in three mistakes are fixed on the next try. On questions they had never seen, 33 of 40 students did better in week four than in week one. One cohort, one exam, no control group: a first signal, not a promise.

Gaplight is the next step and a separate company: the same engine, sold to the people who teach rather than the student, starting with SAT and ACT tutors in the US.

Students already put in the hours. Gaplight finds the gap that costs the most points, plans the next session and forecasts the score.

We sell to the people who teach to the exam. The tutor loads the practice tests they already give, and every student gets their own plan. Before the next session the tutor sees who needs what, and stays in control of the plan. Every parent sees progress they understand, and every result is the tutor’s to show.

The part nobody puts on a landing page: an engine that only knows four-option multiple choice cannot run most real exams. Matching, ordering, a passage you have to read first — those are the questions that carry the marks, and they are the ones that get left out.

So ours asks all of them. That is the difference between saying “any exam” and meaning it. The paper below is a sample we wrote to show the formats; the measured results above come from the real cohort on NMT.academy.