AI-native learning infrastructure
Teaching that adapts to every learner
ClassRoots Edu builds the AI systems behind modern classrooms — a tutor that talks and listens, speech models that hear how a learner actually speaks, and a curriculum graph that keeps every answer on syllabus.
Tracked live
Concept mastery
Sub-second
Voice latency
Standards-aligned
Curriculum grounding
The platform
Six systems, one learning record
Each product solves a distinct problem, and every one of them reads and writes to the same model of what a learner knows. That shared record is what makes the adaptation real rather than cosmetic.
Adaptive Learning
AI Tutor
A real-time conversational tutor that teaches a lesson step by step, listens to the learner, and adapts the path as understanding changes.
Learn moreLanguage Acquisition
Speech & Language
Pronunciation assessment and guided speaking practice that score a learner on accuracy, fluency and completeness — phoneme by phoneme.
Learn moreKnowledge Graph
Curriculum Intelligence
A machine-readable model of what is taught, in what order, and why — the substrate every AI feature reasons over.
Learn moreConceptual Practice
Interactive Simulations
Manipulable maths and science applets the tutor can drive directly, so abstract ideas become something a learner can move.
Learn moreInstitution Platform
School Operations
The administrative backbone — enrolment, content management, scheduling, reporting and exports — that schools actually run on.
Learn moreEducator Workflow
Teacher Tooling
Companion apps that give teachers the operational picture: their day, their classes, and where each learner is stuck.
Learn moreHow the AI teaches
A loop, not a chatbot
Answering questions is the easy part. Teaching means knowing what to say next — and that requires a model of the learner that updates every few seconds.
- 01
Ground the model in real curriculum
Every explanation is retrieved from an approved concept graph aligned to the board and grade the learner is actually studying — not from open web text that happens to sound right.
- 02
Teach, then listen
The tutor narrates a step and then yields. Questions interrupt the lesson, get answered in context, and the session resumes at the exact point it paused.
- 03
Score understanding continuously
Answers, hesitation, speech attempts and applet interactions all update a per-concept mastery estimate. Confidence goes up or down on evidence.
- 04
Choose the next step from evidence
A weak prerequisite reroutes the lesson backwards before it moves on. Demonstrated mastery skips the drill. Nobody sits through content they already own.
- 05
Escalate to a human
When a learner stalls in a way the system cannot resolve, the teacher gets a specific flag — which concept, which attempts, what was tried — not a generic alert.
Then it repeats
The loop runs for the length of the session, the chapter and the school year — which is why the learning record compounds instead of resetting every login.
See the architecture
Responsible by design
AI in a classroom has a higher bar
A confidently wrong answer costs a learner more than it costs an adult with a search engine. These constraints are architectural, not policy documents bolted on afterwards.
Grounded, not improvised
Generated explanations are constrained to approved curriculum. If the graph does not cover it, the tutor says so instead of inventing an answer.
Student data stays scoped
Learner records are tenant-isolated and role-scoped. Access is granted per institution, per role, and is auditable end to end.
Teachers stay in the loop
The system is built to escalate. Where a learner needs a person, it hands over with full context rather than looping them through another prompt.
Measurable, not anecdotal
Mastery estimates, speech scores and interaction telemetry are recorded so outcomes can be evaluated rather than asserted.
Building something for learners?
Whether you are evaluating our platform, exploring a research partnership, or reviewing us for a grant programme, we will get you what you need.