An education that adapts around the learner.
Education intelligence infrastructure that understands what a learner actually knows, decides what should come next, and chooses how it is best taught — from the first years of school to the last.
Access is reviewed individually. There is no public sign-up.
The core promise
One intelligence. Many products. One continuous journey.
Instead of a child adapting to a fixed system, the system adapts to the child — and keeps adapting for as long as they are learning.
Understands the learner
A longitudinal model of what has genuinely been demonstrated — not a grade, and not a guess.
Decides what comes next
Curriculum, mastery, gaps and readiness resolve the next step and can explain why it was chosen.
Chooses how to teach it
Voice, visual work, a project, a simulation or a conversation — whichever suits this concept, this learner, right now.
Measures real learning
Every interaction becomes evidence, evidence is validated, and only validated evidence moves mastery.
One learner
The system remembers, across years and across subjects.
A child's work in a science project, a mathematics explanation and a real-world observation are the same learner seen three ways. Most systems forget that by the end of term.
Longitudinal, not termly
Understanding accumulates. Nothing resets in September.
Cross-subject
A reading difficulty that shows up in word problems is a reading difficulty, and the system can tell.
Cross-context
School, home and real-world learning contribute to one picture.
Precision Learning Twin
A model of what a learner knows — and of what nobody has looked at yet.
The Twin holds concept-level state backed by validated evidence. Its most important state is the honest one: not enough evidence yet.
Evidence before mastery
Mastery changes only through evidence, validation and human review. No score, completion or attendance figure writes to it.
UNKNOWN is not weakness
A concept nobody has assessed is shown as unassessed — never as a low score, never in red, and never as something a child has failed.
Completion is not mastery
Finishing a task, covering a curriculum area, having evidence and having demonstrated learning are four different claims, and the interface keeps them four.
Adaptive learning
A plan that can explain itself.
Each day's plan names why every item is there — the prerequisite that is not secure, the concept due for review, the project waiting on a next step.
Interaction
The right interface for this concept, this learner, right now.
A six-year-old learning quantity and a sixteen-year-old designing an experiment need different interfaces — not different amounts of the same one.
Voice and visual
A learner who cannot yet read completes a full session without reading any interface text.
Workspaces
Whiteboard, project, research, code and simulation — the task decides the surface.
Task-native
Intelligence appears inside the work. The product is not a chat window with education attached.
The Learning Companion
One familiar companion. Many specialised intelligences behind it.
Dozens of specialist teaching agents work behind a single, consistent presence. A learner never meets a picker, a bot name or a handover — and the companion becomes less visible as they grow.
- ≈4–7
Full Companion
Co-present and voice-capable. A child who cannot yet read finishes a whole session without reading a word of chrome.
- ≈8–11
Side Companion
Beside the work, not in front of it. Asks rather than tells.
- ≈12–14
Minimal Mentor
A portrait and a question. The workspace is the page.
- ≈15–18
Intelligent Mentor
Small, serious, and mostly out of the way. Research, projects, code.
The companion is an interface to the intelligence, not the intelligence itself — and it can be turned down to a voice, or off entirely, without losing any part of the product.
Experiences
Learning that happens in the work, not in a conversation about the work.
Projects
Multi-week builds with milestones, a build log and a mentor who asks the hard question before the test run.
Real-world learning
A camera pointed at an ordinary thing becomes an observation, a question and a mission.
Research
Sources, provenance, conflicting claims and citation integrity — with external content treated as untrusted.
Coding
Workspaces, diagnostics and debugging history, in an execution environment isolated from everything.
Simulation
Deterministic, replayable systems where a decision has consequences that can be reasoned about.
Reasoning and debate
Structured argument with no winner declared and no verdict on truth.
Family
What needs attention, and what a parent can actually do.
Not thirty charts. A short answer to the question a parent is really asking, and a specific, safe suggestion for helping.
School
A teacher's time is the scarcest resource in the system.
Teacher surfaces are a working queue, not a dashboard: what needs attention today, what is waiting for review, and which learners need support.
Specialized pathways
The same intelligence, configured for a different life.
A student athlete, a homeschool family, a microschool cohort or a founder programme are configurations of one platform — not separate products with separate records.
Personal School
The flagship individual learning environment.
Early Learner
Voice-first and visual-first, for children who cannot yet read.
Homeschool and microschool
Planning, records, attendance and cohort adaptation.
Athlete scholar
Academic continuity across training, travel and competition.
Future skills
Challenges and demonstrations rather than coverage.
Founder school
Real experiments, with a real ethics gate before anyone else is involved.
Safety and privacy
Built into the architecture, not added at the end.
Every AI interaction passes a safety gateway before it reaches a child. Access is authorized server-side on every request.
No engagement mechanics
No streaks, points, levels or leaderboards. Celebration is for learning, never for showing up.
No surveillance
No facial recognition, emotion detection, gaze tracking or attention scoring. None of it exists in the system.
A companion, not a friend
No guilt, no dependency, no “I missed you”, and nothing kept from a guardian.
Data minimisation
Specialist agents receive only what a task requires. Credentials are references, never values.
Global
Interface language and learning language are different questions.
A parent can read Persian while their child learns in English and speaks Arabic at home. The platform treats those as four separate settings, because they are.
Right-to-left as architecture
Layout, navigation and reading order mirror. Code, mathematics, identifiers and media controls do not.
Scripts, not just translations
Arabic and Persian shape and join correctly — on screen and in exported documents.
A growing registry
The locale registry is a floor, not a ceiling, and the language selector reads from it rather than from a fixed list.
Scale
From one learner to a national system, on one architecture.
Schools, networks and ministries see aggregate patterns with privacy suppression applied and caveats attached — and never a league table.
Multi-campus
Comparison with required caveats. No ranking of schools against each other.
National intelligence
Governed aggregation with small-group suppression that cannot be reversed by arithmetic.
White label
A network can brand the product without being able to change what a warning means.