Proof isn't a slogan here. It's a plan we're building in from day one.
TutoBuddy hasn't run a large-scale evaluation yet, and we're upfront about that. What we can show you is the human need this responds to, the learning-science reasoning behind how Katlego teaches, and exactly how we intend to measure whether it's actually working.
This isn't a hypothetical problem. It's measured, national, and compounding every year a learner goes without support.
81% of Grade 4 pupils cannot read for meaning (PIRLS 2021), a foundation gap that compounds through every subject that follows. By matric, only 1 in 5 learners pass pure mathematics, and only 34.1% attempt it at all (DBE 2025).
This isn't a motivation problem. South Africa has 31,462 vacant funded teacher posts against 12,700 unemployed qualified educators: a resourcing and allocation failure, not a shortage of people who want to teach (DBE / SAHRC, 2024). Globally, 70% of 10-year-olds in low- and middle-income countries are in "learning poverty", unable to read and understand a simple text by age ten (World Bank).
The learners on the other end of these statistics are real people, not data points: a sibling helping a sibling with homework, a learner texting a question late at night before an exam.
Grounded in the most consistently replicated finding in learning science.
Bloom's "2 Sigma" finding
In 1984, Benjamin Bloom published the most cited result in the field: students who received one-on-one mastery tutoring performed, on average, better than 98% of students taught the same material in a conventional classroom, a two-standard-deviation improvement. About 90% of tutored students reached what only the top 20% of a conventional class achieved.
That finding is about human one-to-one tutoring generally. It is not a claim about TutoBuddy's own results, which we haven't measured yet at scale. It's the reason we built Katlego around individual pacing and demonstrated mastery rather than a fixed curriculum schedule, and it's the bar an AI tutor has to clear to be worth building at all.
Source: Bloom, B.S. (1984). "The 2 Sigma Problem." Educational Researcher, 13(6), 4-16.
Not a chat log. A per-topic mastery score that decides what happens next.
Every learner interaction updates a mastery score for that specific topic, not just a record that a conversation happened. The score decides what Katlego does next, so progress is measured, not assumed. Drag the slider to see how.
Remediate. Back to fundamentals before moving on.
Practice. Repeated, varied examples until it sticks.
Challenge. Harder problems to stress-test understanding.
Advance. Move on to the next topic in the sequence.
The evidence roadmap: measuring learning gain, not just usage.
We'd rather show you the plan for earning proof than invent proof we don't have yet.
Foundation & MVP
Curriculum knowledge graph built, mastery engine live, first learners on WhatsApp.
Pilot cohort
3-5 partner schools and NPOs, measuring learning gain on real cohorts, not scale for its own sake.
Productisation
Refining the mastery engine and reporting against what the pilot actually shows.
Evidence engine
Independent evaluation partnership, sized and scoped like education-technology research funders (e.g. J-PAL) structure their grants.
The impact we're aiming at.
A learner who never has to raise a hand in a class of 40 gets the same shot at understanding fractions as one in a class of 15. A parent who can't afford private tutoring gets a second teacher in their child's pocket, for the price of a data bundle. That's the outcome the roadmap above is built to prove, not just claim.
No claims about detecting "learning styles" or other approaches without credible evidence behind them. Just steady, repeatable, mastery-paced practice.
Consent flows and defensible data retention are built directly into the MVP, not added later.
Want to help us build the evidence?
We're looking for 3-5 pilot schools or NPOs, and an academic or evaluation partner to help design an independent assessment of learning gain, not usage metrics.