Method
Why it refuses to just tell you
The default way a language model “teaches” — explain clearly, answer fully, move on — optimises for the wrong thing. It feels helpful in the moment and tends to produce a student who cannot yet do the thing unaided. Every constraint below exists to prevent a specific failure of that kind.
The problem it is built around
A student working with an unrestricted AI assistant can look impressive during practice and then underperform on an unaided assessment, because what was actually practised was prompting the assistant rather than solving the problem. The fluency is real but it belongs to the pair, not the student. A tutor that wants to leave something behind has to withhold, deliberately and consistently.
Tiered help
Assistance is rationed in escalating tiers, and a tier is only reached after a genuine attempt:
- A question back — “what have you tried, and where does it stop working?”
- A pointer to the relevant principle without applying it.
- A hint that names the next move but leaves the execution to you.
- A fully worked analogous problem — never the one you were set.
A complete solution to the actual problem is not on the ladder. If you need one, the tutor is the wrong tool.
Generation before confirmation
Before validating or correcting an answer, the tutor asks for your reasoning or your plan. The act of assembling that explanation is where the understanding forms; the verdict afterwards is close to a by-product. This is why it can feel slower than looking something up, and why the slowness is the point.
Fading
Worked examples matter most for genuinely new material. Once you have demonstrated a step unaided, the tutor stops walking you through it. Re-explaining something already mastered is not thoroughness — it is a missed opportunity to practise.
Verification, not assertion
Quantum mechanics has the useful property that most claims are checkable: the algebra can be expanded, the limiting case tested, the numerical result computed. Where a claim can be checked, it is checked rather than recalled — and genuine uncertainty is stated plainly instead of papered over with confident phrasing.
This is also why the tool is scoped to subjects with mechanical correctness — physics, mathematics, formal logic, programming. In more interpretive subjects the same guardrails would misfire.
Working from your own textbook
A student selects the chapter they are currently studying, and the tutor works from that material so its vocabulary, notation, and problem style match the course rather than a generic treatment. That content comes from the owner's own legally obtained copy of their coursebook, is stored privately, and is never published or served to visitors from this site.
About this site
Quantum Tutor is an independent personal project, built and maintained by one student for their own coursework. It is not a commercial product and not affiliated with any university, publisher, or textbook author. Sign-in is limited to the owner; the pages you can read here are the entire public site.