For educators

Give your students AI that is grounded in your course, not the internet.

Built on how people actually learn. Nothing reaches them until you approve it, and you see where confidence, accuracy and engagement are slipping early enough to act.

QuizzesPractice questions with a confidence rating on every one, so a wrong answer and a lucky guess do not look alike.

LeetLearnA tutor that asks the student questions rather than handing over the answer.

Concept reviewsReadable explanations that stop and check, and re-teach whatever the student missed.

FlashcardsDecks the student answers out loud before the card turns over.

All four are generated from the documents you teach from and stay inside your course. Reviews and flashcards can be authored for the whole class off its weakest topics and published once, and each student then keeps their own place in them.

Why it is built this way

Retrieval practice, being made to pull something out of memory, is the most reliable way to make it stick, and it beats rereading by a wide margin. So every module here is the same idea wearing different clothes.

The quiz is the studyQuestions are how a student works through material, not a test bolted on the end of it.

Reviews stop and askA concept review gates on a short check, and re-teaches what the student got wrong.

Cards demand an answerThe student answers before the card turns, so it is recall rather than recognition.

Confidence comes firstRating certainty before answering is what makes overconfidence measurable instead of a hunch.

The ten methods this is built on, and the one Leeteracy does not do yet, are set out on the evidence page.

What you control

A patent-pending harness, not a single prompt

Generation runs through six stages before anything reaches your queue. We do not publish the sequence, but none of it is a model left to its own devices, and all of it is anchored to your material.

GroundingYour documents are chunked and indexed as vectors with FAISS. Retrieval is scoped to the material you selected, so a question is written from your pages rather than the model’s memory.

HarnessA prompt layer we designed and tune, per discipline and per cognitive level, rather than one instruction handed to a chatbot.

ReviewDeterministic code catches the structural faults. An agent then reads each draft on its own terms and rejects what it cannot defend.

VerificationEvery surviving question is checked back against the passages it came from, and anything that cannot be traced is flagged rather than shipped.

What survives lands pending in your question bank. Nothing skips that step.

Nothing publishes without you

Compiling a quiz scores the assembled set and names what it found. That check reports, it does not block. Publishing is your call, and nothing reaches a student until you make it.

You decide when it goes out

Put a quiz on a date and it publishes that day and emails the class. Recurring series run on the weekdays you pick, and a focus series builds a different quiz for each student from their own weak topics.

And what gets reused

Once students have answered, every question carries its own difficulty and how well it separates those who know the material from those who do not. Building a quiz shows you what it can reuse before it writes anything new.

The two students a grade book will not show you

One is scoring badly and certain they are fine. The other has quietly stopped. Both stay invisible until a result arrives too late to do anything about, so Leeteracy looks for them directly.

Who is confident and wrong

Students rate how sure they are before they answer, so the gap between certainty and accuracy is measured rather than guessed. A student who is confident and wrong does not know to ask for help. This is the view that finds them.

Each student lands in one cell. The label on the left is how they read themselves; the column is how they actually scored.
	Strugglingunder 60	Developing60 to 79	Mastered80 and above
Overconfident	Confidently wrong. The first students to reteach.	Believes the gaps are closed. They are not.	Reads a little high on solid work. Low risk.
Calibrated	Knows they are behind. Needs material, not a warning.	On track, and reading themselves correctly.	Knows it, and knows that they know it.
Underconfident	Behind, and harder on themselves than the score is.	Doing better than they believe.	Doing well and does not trust it. Worth telling them.

Both cut points are set per course. Mastered starts at 80 and Developing at 60 by default, and a student counts as over or underconfident once the gap between their confidence and their score passes 15 points.

Who is drifting

Engagement is scored from what students actually do: work completed, time on task, questions asked of the tutor, whether they come back after a bad result. It surfaces the student who has quietly stopped, well before a grade would have said so.

Bring your own material.

The engine is only as good as what you teach from, which is the point. It writes from your documents, not from the internet.