A system for learning, not a library of lessons

Education should understand learning.

LearnOS turns learner interactions into evidence, evidence into an evolving learner model, and learner state into the next best learning experience.

learner state
evidence
decision
experience
scroll to enter the system
01 / The shift

From delivering content to operating learning.

The unit of intelligence is a decision loop: what do we know, what evidence supports it, and what should happen next?

Traditional LMS / EdTech
CONTENT STUDENT
LESSON QUIZ
SCORE REPORT

Strong at delivery, administration and reporting. Usually weaker at representing why a learner succeeded, failed, forgot, transferred or became independent.

LearnOS
LEARNER STATE EVIDENCE
INTERPRET DECIDE
EXPERIENCE NEW STATE

The platform continuously closes the loop. Each response can change the system's belief and its next action.

02 / The operating architecture

One visible loop. Many governed engines.

The intelligence layer is separated from content delivery. Each engine has explicit inputs, outputs and state changes.

Foundation + LMS
Knowledge architectureGoals · competencies · atomic units · prerequisites
Learning progressionSequences · gates · trajectories
LMS coreIdentity · enrolment · course · roles
Content / experience catalogAssets · activities · blueprints · versions
Assessment / gradebookTests · rubrics · grades · records
LearnOS intelligence coreNext Best
Learning Action
state → evidence → interpretation → decision → experience → response
Defines goals, atomic knowledge units and prerequisites. This is the vocabulary every evidence object points to.
Evidence + decision
Evidence engineEvents → evidence objects → quality
Learner modelMastery · confidence · retention · transfer
Diagnostic + pedagogyHypotheses · interventions · next action
AI experience engineBlueprint → generation → validation
Retention + transferSpacing · retrieval · unfamiliar contexts
Human + explainabilityTeacher loop · trace · override
03 / Aarav — the system in motion

Do not show a score. Show the reasoning.

Aarav is in Grade 3. Goal: understand multiplication and use it confidently in unfamiliar situations.

learner experience / diagnostic probe
event 01 · conceptual signal
3 + 3 + 3 + 3 = ?
12 / correct
event 02 · symbolic signal
4 × 3 = ?
10 / incorrect
learner model / after evidence
Current hypothesis

Concept may be stronger than symbolic retrieval.

The system does not label Aarav “weak at multiplication”. It keeps competing explanations alive until the next evidence separates them.

meaning / groups
86
fact retrieval
49
word problems
78
transfer
38
evidence_id: EV-10482
signal: conceptual_understanding ≠ symbolic_retrieval
uncertainty: 3 competing hypotheses
04 / Evidence is first-class

Every interaction becomes structured evidence.

A response is not a conclusion. LearnOS captures the event, validates it, links it to knowledge and estimates what the event can legitimately support.

01 / EventAnswer, timing, hint, retry, confidence and path.
02 / EvidenceActor, activity, knowledge target, context and result.
03 / QualityReliability, ambiguity, independence and signal strength.
04 / InferenceHypotheses about mastery and misconception.
05 / StateMastery, confidence, retention, transfer and uncertainty.
06 / DecisionNext best learning action.

Architecture note: xAPI uses structured statements for learning events and an LRS for storing and sharing records; profiles can constrain event vocabularies and contextual models can roll raw activity into meaningful learner states.

observations
evidence
learner state
next action
05 / Inside the learner model

State is multidimensional.

“Mastered” is not one number. A learner can understand an idea but retrieve it slowly, solve familiar tasks but fail transfer, or perform correctly while being poorly calibrated.

Mastery

How well the learner can perform the target knowledge or skill.

Confidence

How strong the system's belief is, based on evidence quality and consistency.

Retention

How likely performance is to survive time and spacing.

Transfer

Whether the learner can use the knowledge beyond the practiced surface form.

Uncertainty

What the system does not yet know — and therefore what it should test next.

06 / Aarav decision trace

One response travels through the whole system.

This is the level of explanation the product should expose to teachers, designers and engineers.

01 / EXPERIENCEProblem shown

4 × 3 is tagged to multiplication facts and equal-groups meaning.

02 / RESPONSEAnswer = 10

Response, time, retries and interaction context become an event.

03 / EVIDENCESignal classified

Error is linked to the knowledge unit; evidence quality and hypotheses are stored.

04 / DECISIONProbe, don't punish

A high-information diagnostic is selected instead of generic extra worksheets.

05 / STATEModel updates

Retrieval uncertainty changes; intervention and future schedule change too.

07 / Pedagogical intelligence

The system asks “what should happen next?”

It chooses an intervention from the learner's current state, not simply the next item in a content sequence.

learner state

Meaning: strong
Fact retrieval: uncertain
Transfer: developing
Confidence: moderate

candidate interventions

Concrete groups · array · retrieval probe · worked example · mixed transfer problem

selected action

Array → prediction → map array to 4 × 3 → immediate retry → capture evidence.

08 / AI inside LearnOS

Generative AI is the experience layer.

AI does not decide what learning means. LearnOS supplies the blueprint and validates the generated experience.

StateWhat Aarav needs.
BlueprintPedagogical structure and constraints.
GenerationQuestion, explanation, visual or dialogue.
ValidationCorrectness, difficulty, alignment and safety.
ExperienceRendered for the learner.
09 / The LMS underneath

LearnOS still needs a real LMS.

The intelligence layer does not replace the operational layer. Identity, enrolment, courses, content, assessment, communication and administration remain essential.

01
Identity & access

Accounts, roles, guardians, organisations, consent, permissions and session security.

02
Course & enrolment

Classes, cohorts, schedules, curriculum mapping, assignments and progress.

03
Content & authoring

Lessons, media, question banks, knowledge links, versions and localisation.

04
Assessment & gradebook

Submissions, scoring, rubrics, grading policies and records.

05
Learning intelligence

Evidence, learner model, diagnostics, pedagogy, mastery, retention and transfer.

06
Communication

Notifications, reminders, teacher feedback, parent communication and alerts.

07
Analytics

Learner, class, school and platform views with traceable explanations.

08
Administration

Tenancy, billing, integrations, audit logs, governance and operational controls.

10 / What the user sees

Complexity stays behind the screen.

Aarav sees a clear task. The teacher sees the evidence and the reason for the next action. Engineers can inspect the trace underneath.

aarav / today's learning path
learn
“There are 4 groups with 3 stars in each group. How many stars altogether?”
● ● ●● ● ●● ● ●● ● ●
Prediction: 4 groups of 3 means 4 × 3.
teacher view / why this was selected
Actionable learner state

Meaning strong · Facts developing

WHY NOW
Recent symbolic error + strong equal-groups response.

WHY THIS
Representation bridges meaning → symbol.

WHAT NEXT
Retry independently, then schedule retrieval.

Teacher can accept, adjust, add an observation or override. Human input becomes evidence too.

11 / Feedback + independence

Help should fade as competence grows.

A scaffold that remains forever can create dependence. LearnOS measures what happened after help, not just the final answer.

LEVEL 01Prompt

“What do the four groups represent?”

LEVEL 02Hint

“Try drawing the four groups.”

LEVEL 03Scaffold

Show one group and ask Aarav to complete the rest.

LEVEL 04Explanation

Explain why 4 groups of 3 maps to 4 × 3.

LEVEL 05Independent

New surface form with no support. Stronger evidence.

12 / Retention · transfer · metacognition

Learning is not finished when the lesson ends.

The system tests whether learning survives time, changes context and becomes increasingly self-directed.

RETENTIONAfter a delay, Aarav retrieves 4 × 3 without the array. Performance changes future spacing.
TRANSFERA new surface form uses rows, money, groups or reverse questions to test structural understanding.
METACOGNITIONAarav predicts confidence. The system compares confidence with performance to improve calibration.
13 / Human + machine

The teacher becomes an interpreter.

The system surfaces what matters, why it believes it, what it recommends and where uncertainty remains.

A
Observe

Teacher adds classroom evidence the digital system cannot see.

B
Interpret

System presents evidence and hypotheses rather than pretending certainty.

C
Decide

Teacher accepts, modifies or overrides a recommendation.

D
Learn

Intervention outcomes feed the platform learning loop.

14 / Platform learning loop

LearnOS learns from learning itself.

One learner loop operates inside a second loop that evaluates interventions across cohorts, with privacy and governance controls.

platform intelligenceIntervention
Learning Loop
learner outcomes
aggregate evidence
evaluate policies
improve blueprints
15 / Infrastructure

Start simple. Architect for evolution.

A pilot does not need Kubernetes, a private foundation model or a giant content library. It needs clean contracts between learning engines.

Experience layerWeb/mobile learner UI · teacher UI · admin · accessibility · localisation
Learning servicesDiagnostic · pedagogy · mastery · retention · transfer · feedback · recommendation
Evidence & dataEvent ingestion · evidence objects · learner state · analytics · audit trail
Operational LMSIdentity · enrolment · course · content · assessment · gradebook · notifications
AI gatewayModel routing · blueprint registry · validation · safety · cost controls
Data/platformPostgreSQL/Supabase or equivalent · object storage · queues/events · managed hosting
InteroperabilityxAPI/LRS for learning events · LTI for external tools · analytics contracts
GovernanceConsent · RBAC · privacy · versioning · traceability · model evaluation · override
16 / Dependencies

Build order matters.

The architecture is a dependency graph, not a checklist. A learner model cannot be trustworthy if knowledge units, evidence contracts and event capture are undefined.

Knowledge modelDefines what evidence points to.
Event contractsDefines what gets captured.
Evidence modelTurns events into signals.
Learner stateStores what the system believes.
Decision policySelects the next action.
17 / Planning economics

From pilot to platform.

Planning ranges, not vendor quotes. The major cost is not only software; it is building and validating the learning intelligence.

StageIndicative buildTimeScope
Lean pilot₹8–15L8–12 weeksOne subject · 20–50 atomic competencies · small cohort · core loop
Production SaaS₹35–75L5–8 monthsMulti-role LMS · evidence pipeline · dashboards · AI gateway · production controls
Enterprise / network₹1–3Cr9–18 monthsMulti-tenancy · integrations · governance · analytics · scale
Private AI stack₹2–6Cr+12–24+ monthsDedicated model infrastructure, GPU operations, evaluation and domain controls
LearnOS / The end state

Build the system that can continuously answer:

What does this learner need next — and what evidence tells us that?

Not a bigger content catalogue. Not just a chatbot. A learning system that observes, reasons, acts, measures and learns.

LOOP CLOSED
LEARNOS / VISUAL ARCHITECTURE v3.0Architecture concept · planning artifact