Education for the future,by the future.
Exam revision, rebuilt on the science of learning. Kinetix diagnoses what each student does not yet know, teaches it through interactive lessons, and sustains daily practice through a structured reward system. We are launching with IB Diploma Physics and building for every major secondary curriculum.
Most revision is passive.
Asked how they study, 84% of students named rereading. Only 11% named testing themselves, and just 1% put it first.
Karpicke, Butler & Roediger, Memory, 2009 · 177 US undergraduates listing their own study strategies
You have just read a chapter once. What do you do next?
Most students reread. Only 17.8% tried to recall.
And only 8 of the 101 chose to recall because it helps them remember. Most who test themselves do it to check, not to learn.
The same question, put to 101 US undergraduates: Karpicke, Butler & Roediger, 2009, Table 2. Why students quiz themselves (68% to check what they know): Kornell & Bjork, 2007.
How the research rates ten common techniques
High utility
- Practice testing
- Distributed practice
Moderate utility
- Elaborative interrogation
- Self-explanation
- Interleaved practice
Low utility
- Summarisation
- Highlighting
- The keyword mnemonic
- Imagery for text
- Rereading
Ten common study techniques as rated by Dunlosky et al., Psychological Science in the Public Interest, 2013. Ratings are the reviewers' judgements, not effect sizes.
The loop, on one screen.
A concept of a student's daily view. The diagnostic maps every topic, the gaps surface on their own, and the plan turns them into today's work.
Mastery map 30 topics
- Kinematics82%
- Projectile motion18%
- Momentum71%
- Circular motion29%
- Gravitation64%
- Rigid bodies55%
- Work and energy88%
- Power76%
- Thermal energy69%
- Gas laws81%
- Entropy22%
- Greenhouse58%
- Wave model74%
- Standing waves47%
- Interference33%
- Doppler effect61%
- Diffraction52%
- Refraction79%
- Electric fields41%
- Magnetic fields66%
- Induction15%
- Circuits73%
- Orbits49%
- Fields and potential37%
- Atomic structure84%
- Radioactivity62%
- Fission57%
- Fusion68%
- Quantum ideas26%
- Stars45%
GapBuildingSecure
Today 18 minutes
- Projectile motionInteractive lesson · 6 minGap
- Induction8 retrieval questions · 5 minGap
- EntropyWorked example, then 4 questions · 4 minGap
- KinematicsSpaced review, due today · 3 minReview
This week
×3 multiplier
XP earned per day
Product concept, shown for illustration. Topic scores are invented for the demo.
One adaptive loop.
Every student moves through the same five stages, and each stage produces the data that shapes the next.
1Diagnose
An adaptive assessment maps mastery across the syllabus.
2Plan
Priorities are set by impact and by the time remaining before the exam.
3Learn by doing
Interactive lessons in which students predict, sort and decide.
4Practise
Spaced retrieval practice, reinforced by streaks, XP and multipliers.
5Re-measure
Mastery is reassessed and the plan updated every week.
Grounded in the strongest evidence in learning science.
Seven published findings, redrawn from the original data. Each one shapes a part of the product.
42%of the effort saved by learning a list was gone within a day (58% savings at 20 minutes, 34% at one day).
Memory fades fast, then levels off.
How much less effort it took to relearn a list after each delay, in Ebbinghaus's 1880 study and a 2015 replication. Lower means more was forgotten.
In the product Kinetix is built to schedule the first review before that drop, then space the next ones further apart.
Show the data
+14 ptsmore recalled a week later after one recall test instead of a reread (56% vs 42%).
Testing beats rereading once time passes.
Students read a passage, then either restudied it or took one recall test without feedback. Restudy led at 5 minutes; testing led at 2 days and a week.
In the product Every Kinetix lesson is designed to end in retrieval questions, not a reread.
Show the data
21 dayswas the best gap between sessions for a test a year away. For a test a week away, it was one day.
The best gap between study sessions grows with the time until the test.
Recall of facts studied in two sessions, by the gap between them and by how long until the final test. Larger dots mark the best gap tested.
In the product Kinetix is built to set review gaps from the exam date, widening them the further away it is.
Show the data
1.5×as many students failed with lectures as with active learning (33.8% vs 21.8%).
Fewer students fail with active learning.
Failure rates in undergraduate STEM courses, across 67 studies.
In the product Kinetix lessons are designed so the student acts at every step.
Show the data
2×the conceptual gains in interactive physics courses (0.48 vs 0.23).
Interactive physics courses doubled conceptual gains.
Average normalised gain on mechanics tests, 62 courses and 6,542 students. Whiskers show the spread across courses.
In the product Kinetix starts with IB Physics, the subject this evidence comes from.
Show the data
+23 ptson a test a month later when practice mixed problem types (60.7% vs 37.6%).
Mixed practice lifted scores by 23 points.
Grade 7 maths in 54 classes, 787 students, on an unannounced test a month later.
In the product Kinetix practice sets are designed to mix topics instead of drilling one at a time.
Show the data
52%lower prediction error than SM-2, the classic flashcard scheduler (log loss 0.346 vs 0.722).
The scheduler Kinetix uses predicts memory most accurately.
How well each scheduling model predicted recall across about 350 million Anki reviews from 9,999 users.
In the product Kinetix schedules reviews with FSRS-6, the most accurate model in the benchmark.
Show the data
Also relevant: classroom quizzing raised achievement by g = 0.50 across 222 studies (Yang et al., Psychological Bulletin, 2021), and game elements improved learning outcomes by g = 0.49 (Sailer & Homner, Educational Psychology Review, 2020). All results are drawn from published research, not from Kinetix, which has not yet launched.
Designed for learning, not conversation.
Recent trials of AI tutoring show mixed results. A purpose-built tutor at Harvard more than doubled median learning gains, while general-purpose chat tools have shown little benefit and, in some cases, lower grades. Outcomes depend on design. Kinetix is built around structured practice and retrieval rather than open-ended chat.
Kestin et al., Scientific Reports, 2025 · Oreopoulos & Low, NBER, 2026 · Liu et al., EdWorkingPaper, 2026
A focused entry into a large market.
We begin with IB Diploma Physics, then expand across the IB, A-level and AP, and national examinations.
3,243,979
AP students, 2024–25: where we expand next
906,425
A-level entries, June 2026: alongside AP
US$12.1B
online tutoring market, 2025 (analyst estimate): the wider category
Built by people who know the problem.
Kinetix was founded by students who have lived the problem: preparing for high-stakes final exams with tools that were never designed around how people learn. We are now speaking with early investors.
Scheduling link coming soon.