Learning that adapts to you — not to a fixed syllabus.
The best result in education research is one-to-one, mastery-based tutoring — it can lift an average student close to the top of the class. This is our plan to bring that within reach at scale: teach against what you actually know, prove real mastery before moving on, and pull in a human exactly when you need one.
Rotational Motion · per-skill mastery
Next: patch “Rolling without slipping” first
One syllabus, one pace, for everyone
A fixed sequence of lectures assumes every student arrives with the same gaps. Most exam marks are lost not on the current chapter but on an unremediated prerequisite from months ago — the one thing the class never went back to fix.
Your gaps decide what comes next
We track your understanding at the level of individual skills, find the weakest one blocking your progress, and teach that first. You advance when you have demonstrably learned something — not when a timetable says so.
The adaptive learning loop
Every topic runs this loop, one small skill at a time. It isn't a fixed sequence — each step is a decision driven by what you've just shown you can (and can't) do.
- 1
Prerequisite check
Diagnose the skills this topic builds on — and patch any gap before going further.
- 2
Set the starting point
Estimate how well you already know each skill — a live belief, not a single grade.
- 3
Teach at your edge
Worked example → faded steps → solo, held near an 85% success rate.
- 4
Retrieval practice
You do the work. Hints on demand; every wrong option maps to a misconception.
- 5
Prove mastery
Hard items, speed, low hint use, honest confidence — and a check days later.
- 6
Spaced review
Mastered skills return on an expanding schedule, so they stick instead of fading.
When you get stuck
A repeated misconception triggers targeted re-teaching of that exact error. A plateau or mounting frustration routes you to a human mentor — with the AI handing over precisely where you're stuck.
Forgetting re-opens the loop
If a delayed check shows a skill has slipped, it quietly comes back into rotation. “Learned” means it lasts — not that you passed it once.
A topic is a set of skills — we track every one
A single “topic score” hides where you're actually weak. We break each topic into knowledge components and keep a live estimate — a probability you know each one — so remediation targets the right thing.
Topic: Rotational Motion
P(known) per skill · illustrativeA single weak skill — here, rolling without slipping — blocks the chapter even when the average looks fine. The engine patches that weakest link first, in the right order, instead of re-teaching what you already know.
“Learned” is more than a passed quiz
One 20-question quiz can be beaten by luck or memorised patterns. A skill only counts as mastered when every one of these signals agrees.
Accumulated evidence
A high, stable belief you know it — built over many attempts, not one run.
Hard items, not easy ones
Correct near the top of the difficulty range, not just the gentle ones.
On your own
Recent successes reached without leaning on hints.
At exam speed
Fluent, timed responses — the real exam is against the clock.
Honest confidence
Your certainty matches your accuracy — no confidently-wrong blind spots.
Survives a delay
Passes a spaced check days later. This is what separates learned from crammed.
Mastered
Only when all six signals agree. Miss one — it survives practice but not a delay — and the skill stays open for more work.
AI handles the volume. Humans handle the moments that matter.
The scarce resource is human attention, so we spend it where it changes your trajectory — and let AI do everything it does well, instantly and for free.
EduSaathi AI
Always onThe default tier: instruction, hints and step feedback, misconception-tagged remediation, and spaced review scheduling. Instant, patient, and infinitely scalable — it carries normal practice and exam-style drilling.
Human mentor — async & cohort
On triggerWhen you plateau, keep hitting the same misconception, or your confidence stops matching your results, a mentor steps in — often in a small-group session with others stuck on the same thing. The AI pre-briefs them on exactly where you are.
Live 1:1 — rare, high-value
ReservedSaved for conceptual breakthroughs, real frustration, or when you simply ask. Whatever the mentor concludes flows back in, so the AI picks up coherently afterward.
The guardrail: we escalate on how you feel and how hard you're working — not only on whether you got it right. A frustrated student who is technically passing still needs a human; a calm student who missed one item does not.
Every choice traces to evidence
None of this is a hunch. Each principle below has a large, repeated effect on how much students actually learn.
Mastery learning
Advance on demonstrated learning, not seat time.
The 85% edge
Practice pitched just beyond your current ability — where learning is fastest.
Retrieval practice
Doing beats watching by a wide margin. Practice is the product.
Spacing & interleaving
Expanding intervals and mixed problems build durable, transferable memory.
Manage cognitive load
Worked example → faded → independent, as your skill grows.
Formative feedback
Every wrong answer names the misconception and what to do next.
Metacognition
Track confidence against accuracy to surface blind spots.
Motivation, done right
Real progress and human connection — not streak-bait and leaderboards.