A tutor that remembers your mistakes can help you fix them. A tutor that remembers them where you cannot see is something else. We think memory is the most useful and the most delicate part of an AI tutor.
Good teachers remember. They know which student always mixes up capital and revenue expenditure, and which one gets the concept but loses marks on presentation. That memory is what turns one lecture for sixty students into sixty slightly different conversations. An AI tutor can do the same, and it can do it for every student, every evening. But a record of everything a student has struggled with is also a sensitive record, and how it is kept matters.
Without memory, every session starts from zero. The student re-explains where they are, the assistant re-discovers the same gap, and nothing compounds. With memory, the tutor can do the things that actually help before an exam:
In Teach-Back, the assistant remembers what each student mixes up and schedules ten-minute revisions on the phone, in the student’s language. Faculty get a gap heatmap across the cohort. That is the point of memory: gaps closed before the exam instead of found at results.
The same record, kept out of sight, changes the relationship. A student who suspects that every wrong answer is being stored somewhere will stop experimenting, stop asking “stupid” questions and start performing for the record. That defeats the purpose of a Socratic tutor, which depends on the student being willing to be wrong out loud.
Hidden memory also tends to hold more than it should. Conversations wander. A student may mention a family situation, an ID number or a one-time password while asking about something else, and that detail has no place in a learning record.
There is a quieter risk too. A record that only the institution can see starts to look like an assessment. Once students believe their practice sessions feed into how they are judged, practice stops being practice. We think the line should be explicit: memory exists to help the student revise, and faculty see patterns across the cohort so they can teach better, not a running file of every slip a student made while learning.
| Principle | What it means | How Teach-Back applies it |
|---|---|---|
| Visible | The student can see what is remembered | Memory is shown to the student |
| Editable | The student can correct or remove items | Students can edit it |
| Erasable | The student can start over | Students can wipe it |
Alongside these, Teach-Back is private by default: no ID numbers or OTPs go into prompts. And because it runs on managed cloud, in an institution’s own VPC or on-premises, with India data residency when required, the college decides where student data lives.
If you are evaluating an AI learning tool, we would suggest putting memory on the list of things you ask about directly, next to accuracy and academic integrity:
Your institution’s data-protection obligations should be confirmed with your own legal adviser; these questions are about whether the tool is designed to respect them. For the wider policy questions AI raises on campus, see our AI and academic integrity checklist for colleges, and for the tutoring approach itself, the guide to a Socratic AI tutor for commerce students. More on the Teach-Back guides page.
It helps, because it can schedule revision on the exact gaps before a test. But students should be able to see, edit and wipe what it remembers.
It remembers what each student mixes up and uses that to schedule short revisions. The memory is visible to the student, who can edit or wipe it.
No. It is private by default, with no ID numbers or OTPs in prompts.
See it on your own data. Teach-Back — Learn it, then teach it back. Book a 30-minute working session with an engineer.
