The Academy That Remembers You

Personalisation here is not a recommendation widget. It is three distinct systems, each remembering a different kind of thing about you.

The memory layer: a record of you as a learner

Every meaningful thing you do in the academy emits a memory event: the pre-test you guessed through, the Feynman attempt that passed on the third try, the free recall that kept missing the same concept, the debate, the exam, the mock, the research session, even how well you stuck to your study plan. That stream is the raw material.

Then, periodically, a consolidator does what your own memory does overnight: it folds the events into durable facts ("struggles with limits at infinity," "strongest in the morning sessions," "passed the thermodynamics gate after revisiting entropy") and retires facts that new evidence contradicts: superseded, not silently deleted, so the record stays honest. Above the facts sits a compact learner profile: the running picture of how you study.

This is what makes Lura feel like she knows you, because she does. When she plans your week, nudges you toward a topic, or explains why something is due, she is recalling your facts, not applying a template. And the memory is yours to see: the Memory section in Settings shows what the academy holds and lets you manage it.

FSRS-6: fitted to your forgetting curve

Spaced repetition (Cepeda et al., 2006) lives or dies on one question: when will you forget this? Norudit schedules every card with FSRS, built on the memory model of Ye et al. (2022), and personalises it in two layers.

By default, your history shapes every interval. FSRS models each card's memory with stability and difficulty, updated on every single review from your grades. Struggle with a card and its intervals compress; find it easy and they stretch. Two students studying the same deck end up on completely different schedules within a week, because the model is tracking two different memories.

Then, one tap fits the whole algorithm to you. Once you've logged about 400 reviews, the academy can run the full FSRS optimiser on your review log (the same optimisation used by the algorithm's own researchers) and fit all of FSRS-6's parameters to your measured forgetting curve (Ebbinghaus, 1885), scored against your actual recall record. From then on, every interval in every deck is computed with weights trained on you, not on a population average. (Under 400, the fit wouldn't have enough signal to be honest, so it waits; the research defaults are the safe start.)

Combined with concept-varied cards, the effect is a review system where both halves are personal: when a concept returns is fitted to your forgetting curve, and what returns is a fresh variation that only understanding can answer.

BKT: a mastery ledger per concept

The third system tracks not memory but mastery. Every graded signal in the academy (the gate, free recall, exams, mocks, each review answer) is an observation on a specific concept, and Bayesian Knowledge Tracing (Corbett and Anderson, 1994) folds them into a running per-concept estimate of how well you actually know it. That ledger is what makes the rest of the academy pointed instead of generic: the Smart Scheduler puts your weakest concepts first, reviews surface them sooner, and mock papers deliberately re-test them in fresh scenarios.

Why this is the flywheel's engine oil

The learning flywheel closes because the system remembers you between sessions: the exam's diagnosis lands in the mastery ledger, the mastery ledger steers the scheduler, the scheduler feeds the reviews, the reviews refine both your forgetting curve and your mastery, and Lura narrates the whole loop back to you in plain words. Take the memory away and it's a conveyor belt again, restarting cold every morning.

The memory layer, the fitted scheduler, and the mastery ledger are part of everything else in the academy. Start your first class →

Sources

• Cepeda, N.J., Pashler, H., Vul, E., Wixted, J.T. and Rohrer, D. (2006) 'Distributed practice in verbal recall tasks: a review and quantitative synthesis', Psychological Bulletin, 132(3), pp. 354–380.
• Corbett, A.T. and Anderson, J.R. (1994) 'Knowledge tracing: modeling the acquisition of procedural knowledge', User Modeling and User-Adapted Interaction, 4(4), pp. 253–278.
• Ebbinghaus, H. (1885) Über das Gedächtnis: Untersuchungen zur experimentellen Psychologie. Leipzig: Duncker und Humblot. Translated as Memory: A Contribution to Experimental Psychology.
• Ye, J. et al. (2022) 'A stochastic shortest path algorithm for optimizing spaced repetition scheduling', in Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York: ACM. (Extended version in IEEE Transactions on Knowledge and Data Engineering, 2023.)

Read more

• How Norudit's flashcards rebuild themselves every time so you can't just memorise the card: The Deck That Rewrites Itself
• How Norudit plans your actual study week like maths, not a guess: The Scheduler: A Study Plan Run by an Algorithm, Not an AI Guess