Most maths platforms drill what a child got wrong this week. We find the earlier gap that’s causing it.
Existing platforms are excellent at delivering practice on the topic a child is currently struggling with. But for children in care, that topic is rarely the actual gap — it’s the symptom of another gap, left by disruptions to their schooling months or years earlier. STEMeUp is built to find that earlier gap, fill it, and unlock what they’re ready to learn next.
Symptom vs root cause
UnderstoodRoot GapBlocked By GapGoal
Imagine a Year 10 student stuck on a linear equation. The teacher sees “struggling with algebra” and assigns more practice. But that’s not where the problem started. Eighteen months ago the student changed placements and missed a couple of weeks on negative numbers — and every equation since has been tripping over that same gap, without anyone catching it. More algebra won’t help. What helps is finding the thing underneath the algebra.
That’s what our engine does. It works backwards from where a student is visibly stuck to the earlier gap that’s actually causing it, so we can fix the real problem instead of drilling the symptom.
The learner journey at STEMeUp
Finding one gap is just the start. From there, the platform keeps working out what a student knows, what they don’t yet, and what they’re ready to learn next based on what they already understand. Each concept they master opens up the ones that build on it, so there’s always a sensible next step. It carries on like this, concept by concept, adapting as the student goes, until they reach their goal.
Gap mapping
Year 9
The engine assesses every concept and finds the gaps.
→
STEMeUp Tutoring
Year 9–10
Foundations first — the earliest gaps fill before anything new.
→
STEMeUp Tutoring
Year 10–11
Each mastered concept unlocks the ones built on it.
→
Grade 5+ achieved ✓
Year 11
The whole map mastered — goal in hand.
Gap Being AssessedGap LocatedMastered
The three things that make it work.
01
An AI tutor that adapts to the moment.
Across the three-year window from Year 9 to GCSE, the tutor adjusts mode and difficulty to the learner — teaching a concept, asking the learner to teach it back, or offering an easy win when the right move is to rebuild confidence before pushing on. Every mathematical output is checked for accuracy before it reaches a child.
02
A knowledge-gap mapping engine.
GCSE Maths is a web of connected concepts, each one resting on those below it. We map every learner precisely onto that web — concept by concept — so when a child gets stuck, we can tell the difference between a topic they haven’t met yet, one they’re actively struggling with, and one whose root cause sits several steps upstream.
03
A data layer that works for everyone.
The platform is built around what works for the people supporting the child. An up-to-date picture of each learner’s engagement, time on task, progress, and gaps — written in language that drops straight into PEP meetings, statutory reporting, and Ofsted evidence. Only three clicks away from an individual or cohort-level report. No setup, no extra admin. It’s a resource to use if and when you need it.
Chatbots hand over answers. Our tutor never does.
The research below shows why that choice matters: when a tool gives students the answer, they stop doing the thinking that makes learning stick — and can end up scoring worse in the exam than students who never used it at all. Our tutor breaks each problem into steps, works through examples, and asks the learner to explain the idea back.
Who we’re building this with.
We build STEMeUp with the people who use it:
Virtual School Heads
PEP Caseworker
Designated Teacher
Carer
Learner
Learners at every point of their GCSE journey — whether the exam is three years away or three months.
Carers whose continuity around the child matters as much as anything the platform does.
Designated Teachers who want a clear picture of each child and their areas of need.
PEP Caseworkers who want evidence of progress that doesn’t take extra time to assemble.
Virtual School Heads who need cohort-level visibility into the incremental progress children make — a granularity that attendance and grades alone cannot offer.
We’re rolling out across the North East first, beginning with Durham, before scaling nationally.
Built to be safe and secure. The platform meets UK GDPR requirements and is built to the Cyber Essentials standard, with safeguarding embedded throughout.
Interested in piloting with us?
We’re selecting a small group of schools and local authorities for our first pilot cohort. Get in touch to find out more.
Ünal, E., Kaya, M., Uzun, A. M., & Erdem, C. (2026). A Second-Order Meta-Analysis on the Effects of Artificial Intelligence Applications on Student Outcomes. Journal of Educational Computing Research. doi.org/10.1177/07356331261424767
Kestin, G., et al. (2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15(1). doi.org/10.1038/s41598-025-97652-6
Escueta, M., Nickow, A. J., Oreopoulos, P., & Quan, V. (2020). Upgrading Education with Technology: Insights from Experimental Research. Journal of Economic Literature, 58(4). doi.org/10.1257/jel.20191507
Ma, W., Adesope, O. O., Nesbit, J. C., & Liu, Q. (2014). Intelligent tutoring systems and learning outcomes: A meta-analysis. Journal of Educational Psychology, 106(4), 901–918. doi.org/10.1037/a0037123
VanLehn, K. (2011). The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems. Educational Psychologist, 46(4), 197–221. doi.org/10.1080/00461520.2011.611369
Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2024). Generative AI Can Harm Learning. The Wharton School Research Paper. ssrn.com/abstract=4895486
Kosmyna, N., et al. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. MIT. arxiv.org/abs/2506.08872
OECD (2026). OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education. OECD Publishing, Paris. doi.org/10.1787/062a7394-en
Wang, R. E., Ribeiro, A. T., Robinson, C. D., Loeb, S., & Demszky, D. (2024). Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise. arXiv:2410.03017. arxiv.org/abs/2410.03017