01
A persistent record of what you avoid
Recognition and production are tracked separately. Knowing a structure and reaching for it unprompted are different things, and only one of them shows up in a test score.
A language learning company
AI can handle the practice. Teachers should do the rest.
Most tutoring time still goes to work a machine now does better: generating exercises, explaining grammar, drilling vocabulary. We build the learner model that lets a teacher spend the hour on everything else.
One learner model
Every hesitation, avoidance and repeated error, held in one place.
Practice between lessons
Chat, listening, reading and review, generated from real mistakes.
A brief before the lesson
The tutor arrives knowing what to test and what to skip.
The problem
Language tutoring is still organized around delivering content.
- textbook
- chapter
- vocabulary & grammar
- exercises
- homework
- next chapter
AI is unusually good at most of what made that model necessary: generating exercises, adapting difficulty, producing examples, explaining a mistake a second and third way, simulating conversation, remembering recurring errors.
Very few tutors have moved, and the reason is not that they haven't noticed. Books solve curriculum design, sequencing, exercise generation and homework in advance. A tutor can open the book at the page the student stopped on and teach several students a day with almost no preparation. Designing an individual curriculum every week costs far more.
So the opportunity is not replacing the textbook with a chatbot. It is giving every tutor the infrastructure to run an individualized curriculum without doing the planning by hand.
The shift
From content delivery to diagnosis and intervention.
- 01Observe the learnerAI, between lessons
- 02Infer the current gapsLearner model
- 03Choose the next useful challengeTutor decides, AI proposes
- 04Let the learner attempt itLesson or practice
- 05Give the minimum useful interventionTutor
- 06Update the model, repeatAI
That is a different job, and a harder one. It also makes the human more valuable, not less. Their scarce skill was never producing another worksheet. It is noticing that you technically know a structure but avoid it when speaking, that your pronunciation is understandable but unnatural here, that this is a moment to push rather than explain.
What we're building
A learner model both the AI and the human can act on.
Generating exercises is no longer the hard part. Everyone can do that. The hard part is knowing the learner well enough that both sides know what should happen next.
02
Practice generated from your own mistakes
Conversation, listening, reading and spaced review, built around the errors that actually recurred this week rather than the next chapter in a book.
03
Preparation the tutor did not have to do
A brief before the lesson, a suggested shape for the hour, and homework written from what happened in it. The planning cost that kept tutors on textbooks goes away.
Five minutes before the lesson, the tutor gets this.
Not a transcript and not a dashboard to interpret. A short brief that says what the week showed, and one suggested shape for the hour. The tutor overrides any of it.
Afterwards, what happened in the lesson becomes the following week's practice.
This week, independently
94 minutes of practice. Vocabulary retention is strong — do not spend lesson time on it.
Production gap
把 encountered 17 times, produced correctly 11 of 15. Recognition strong, spontaneous production weak.
Recurring error
过 and 了 confused when describing past experiences.
Pronunciation
zh / ch distinction slipped in 6 conversations.
Suggested lesson
15 minutes on travel to elicit 过. Introduce an unexpected problem that needs 把. Correct pronunciation only where meaning breaks. Close with free conversation.
Division of labor
This is AI-augmented tutoring, not an AI tutor.
The split is not a compromise. Each side gets the work it is actually good at.
AI
Human
The consequence
The human hour becomes a premium resource, deliberately spent on nothing a machine could have done. Three parties gain at once: the learner gets personalization, the tutor gets preparation and student tracking for free, and the platform sits between independent practice and human instruction.
In brief
The facts.
- What mantiks is
- A language learning product built on a persistent learner model, used by learners on their own and by tutors in lessons.
- The bet
- Exercise generation is a commodity. A learner model good enough for a human teacher to act on is not.
- Who it is for
- Learners who already practice daily, and the tutors they pay for the hour that practice cannot replace.
- Surfaces
- Chat with a tutor model, listening, reading, quizzes and spaced cards on web and mobile, plus a tutor view.
- Where we are
- Building. Early access opens language by language.
- Contact
- hello@mantiks.ai
We're building it now.
Learners get early access as languages come online. Tutors join a small first cohort and help shape the brief.