Every course becomes a knowledge graph. Every student walks their own path through it — by talking, not by reading.
What came out of the first conversations with a university — and what the system does about each one.
Every document that goes in is broken down into knowledge components and the prerequisites between them — including links that run across documents, not just within one. The syllabus stops being a pile of PDFs and becomes a structure that can actually be reasoned about: what comes first, what depends on what, where a gap really is.
Because the graph knows what depends on what, and the system knows what you have already demonstrated, no two learners get the same sequence. You are never quizzed three concepts past where you are, and never marched through something you already own. The course is the same; the route through it is yours.
So the teaching happens in conversation. The tutor explains, asks back, corrects, and lets you arrive at it by talking — with the course material underneath as the source rather than as the interface. The textbook still decides what is true. It no longer decides how you get at it.
Finding out where someone stands does not have to mean a multiple-choice battery. The placement check probes along the backbone of the subject instead of testing everything, and the tutor reads what you know from how you discuss it — so ordinary conversation counts as evidence, not just answers typed into a form.
And the point of all of it: learning that people come back to. Not a duty that gets postponed until the week before the exam — a conversation worth having again tomorrow. That is the design goal every one of these choices serves.
Drop in lecture notes, PDFs or a whole git repo of course content. It is indexed into a searchable knowledge base — not a generic model's guesswork.
The material is broken into knowledge components with prerequisites between them, so the tutor knows what has to come first — and what you are not ready for yet.
A Socratic chat tutor grounded in your documents. It does not hand you the answer — it makes you produce it, and it says so when you are wrong.
Everything you have shown you know comes back for review on an FSRS spaced-repetition schedule, so mastery is measured over weeks, not in the moment.
Markdown and PDF via upload or a watched folder, or synced automatically from git repositories. Re-ingest is idempotent — only what changed is reprocessed.
An LLM proposes the teachable units and their prerequisites. A teacher reviews and approves them — the map is not left to the model alone.
Chat with a tutor that answers only from the indexed material, in teach mode or practice mode, with full markdown and LaTeX rendering.
Practice questions are generated per knowledge component and graded, in an order that follows the structure of the material rather than random shuffling.
An interactive knowledge map and statistics show, per subject, what is mastered, what is shaky and what has not been touched — with review scheduled accordingly.
Answers come from retrieval over your own documents, so the tutor teaches the course you are actually taking — not the internet's average version of it.
Knowledge components, prerequisites and where the material starts are approved by a human. The model proposes; the teacher decides.
Bring your own LLM API key, per user. Content and progress are isolated per user and enforced in the database itself, not just in the UI.
The interface and the teaching are localized — you can learn hard material in your first language while your English is still catching up.
The idea is not new here. It was built in 2024 as a grounded chat tutor and piloted for a full academic year — at Corvinus University of Budapest, in Solana Superteam's educational tracks, and in the founder's own teaching at Óbuda University. That year proved the premise and exposed its limit: a chat that answers is not a curriculum. Stage II is the rebuild that answers it, with a knowledge graph underneath.
a year of piloting with students
graph engine, mastery tracking, courses
running at tutor.trypolearn.com
Bring your own course material and find out what you actually know.