Eight research-backed pillars from cognitive science. A learning system designed pedagogy-first, not AI-first.
Most AI tutoring products start with a language model and add some education on top. PrepGraph does the opposite. We start with the cognitive-science research on how students actually learn and retain — and use AI to deliver that pedagogy at scale.
A ChatGPT-style chat lets the language model freely decide what to say. That produces fluent answers but doesn't compound into real learning. PrepGraph keeps the what grounded in research — and uses natural language only for the how.
"The value of intelligent tutoring systems depends on design, context, and evaluation quality."
— Nature (2025), A systematic review of AI-driven intelligent tutoring systemsEvery claim PrepGraph makes is traceable to one of these eight pillars. Every pillar is traceable to a citation we can defend in front of your academic dean, your IT review, or a curious parent.
You don't advance until you've actually shown you can do it.
PrepGraph is mastery-based, not exposure-based. You advance when you've demonstrated you can apply a concept — not because you sat through a video. No padding stats with easy questions you already know.
Bloom's mastery learningWe make you pull knowledge out, not just put it in.
Sessions begin with cold retrieval of concepts from earlier work. Mid-session checks. End-of-session mixed recall. Delayed retrieval across days and weeks. The EEF rates retrieval practice as one of the highest-impact, lowest-cost interventions in education research.
EEF — "Why bother with retrieval?"Concepts come back at the right time. If you start forgetting, they come back sooner.
Blocked practice early, mixed practice in the middle, time-bound interleaved practice at exam stage. Review timing adapts to each student's forgetting pattern, not a one-size schedule. The Nature 2025 systematic review of intelligent tutoring systems calls this the single most-overlooked feature of generic AI-tutoring tools.
EEF Cognitive Science research agenda · Nature (2025)We never explain more than your working memory can hold.
Length and complexity of explanations adapt to where the student is. Confused → switch to a worked example. Intermediate → guide with questions. Advanced → challenge with variation. Long explanations to struggling students are a known failure mode of generic AI tutors — we design against it.
Sweller / cognitive load theoryYou learn what kind of student you are.
After sessions, short reflection prompts: what was hardest? Was it the concept, the math, or how the question was phrased? Your dashboard tracks not just "% correct" but consistency, confidence, and the kinds of mistakes you tend to make. EEF rates metacognition as a high-impact, low-cost approach.
EEF — Metacognition & Self-Regulated Learning guidanceThe tutor will not hand you the answer.
PrepGraph asks you to attempt first. It nudges with hints when you're stuck, then with guiding questions, and only explains in full at the end. Brookings' analysis of GenAI tutoring identifies this "teach, not tell" pattern as the single highest-leverage design principle.
Brookings — design principles for AI tutoringA wrong answer is data. We learn from it.
When you get a question wrong, PrepGraph identifies what kind of mistake it was — a concept gap, a slip, a misreading — and chooses the next problem to address that specific issue. Over weeks, your weak-areas list visibly shrinks. This is the difference between a question bank and a learning system.
Cognitive-science research on error analysisIn schools, AI is the layer beneath the teacher.
Teachers see class misconception heatmaps and get specific intervention recommendations. Parents see weekly evidence-based reports. Students get a Socratic tutor. None of this replaces a human teacher — Brookings' GenAI tutoring research is explicit that the strongest evidence comes from hybrid models where AI handles personalisation and humans handle judgment.
Brookings · UNICEF (2025) AI-and-children guidanceThe pillars above describe what we believe about learning. The capabilities below describe what each of those principles looks like in practice.
Content aligned to CBSE/NCERT and the major exam syllabi for JEE, NEET, SAT, and ACT — ICSE and state boards are on our roadmap. Teachers can layer their school's scope and sequence on top.
PrepGraph builds a personalised picture of each student — what they know, where they get stuck, how long they've held onto concepts, their pace, and where they are against their goal. Not a single "% mastered" number.
Every session adapts. What to teach next, when to remediate, when to bring a concept back, when to increase difficulty, when to slow down — all driven by the learning data, not a fixed playlist.
The natural-language tutor that delivers the personalised plan. Designed around teach-not-tell. Asks you to attempt, hints when stuck, guides through the steps, and explains in full only at the end.
You progress when you've shown you can do it. Concepts you've mastered come back later for retention checks. The mechanism that turns "I learned it" into "I remember it on exam day".
Anonymised learning analytics, pre-test/post-test gains, retention measurement, exam-readiness trends. The evidence base that lets us publish outcomes, not just claim them.
We hold ourselves to honest scope. Today, every PrepGraph student gets a Socratic AI tutor, mastery-based progress, exam-readiness projection, and weekly parent reports. Some capabilities are still being expanded — we say so explicitly rather than promising what we can't yet deliver.
Free to start. Take the adaptive diagnostic. Feel a Socratic tutoring session. Watch the spaced-review schedule fill in over your first week.