The Visibility Gap
Schools cannot see which AI tools students actually use, for what, or how often — most tools exist on paper, not in practice.
Loading…
AIforAll SLM · Early Student Support System
AI shouldn’t be a privilege.Neither should safety.
A purpose-built Small Language Model that makes safe, responsible AI adoption in K-12 schools measurable, proactive, and human-governed — piloted across the UAE and India.
- Safety-First SLM
- UAE + India Pilot
- School-Governed Data
Why now
AI is already in classrooms. Safeguards aren’t.
Students in the UAE and India are adopting general-purpose AI tools faster than schools can govern them. Written policies describe the right behaviour — but no system watches, measures, or intervenes in real time.
The Wellbeing Gap
Distress signals inside AI conversations — isolation, self-harm cues, unhealthy dependence — go undetected until a crisis surfaces elsewhere.
The Governance Gap
Data residency, consent, and security rules differ across the UAE and India — most tools are built for neither jurisdiction.
The solution
Introducing AIforAll SLM.
Not another general chatbot dropped into classrooms — a small, domain-tuned language model designed from the ground up to be bounded, observable, protective, and governable by the school itself.
Bounded
Scoped to curriculum, age band, and approved use-cases only — no open-ended general chat.
Observable
Every interaction is logged and scored against safety and integrity signals, not silently discarded.
Protective
Detects wellbeing risk signals and routes them to a human — never an automated diagnosis.
Governable
The school, not a vendor, controls data residency, retention, and policy configuration.
The framework
Four pillars, one predictive core.
Proactively measuring how AI is actually used — before a small issue becomes a serious one. Each pillar feeds a Predictive Risk Engine at the centre.
Predictive core
Predictive Risk Engine
The core that reads every pillar’s signals and scores risk in real time.
Tools
Approved AI-tool registry, access levels by age band, usage visibility.
AI & Data
Model behaviour limits, data minimization, consent, retention rules.
Security
Encryption, jurisdiction-local hosting, access control, incident response.
Process
Escalation workflows, review cadence, stakeholder accountability.
Protecting students
How the predictive safety layer works.
Signals are read, scored, and routed to a human — the system never diagnoses or decides alone.
Signal Ingestion
Interaction patterns, language cues, and usage timing (e.g. distress language, late-night spikes) are read in real time.
Risk Scoring
Signals are scored against calibrated wellbeing & integrity thresholds — privacy-preserving, no raw content stored longer than needed.
Tiered Escalation
Low tier: self-help resource shown. Mid tier: counsellor/teacher alerted. High tier: parent + safeguarding lead notified immediately.
Human-Led Response
A trained adult always makes the final call — the model surfaces concern, people decide and act.
Dual-market rollout
One core model, two regulatory realities.
The same ESSS core adapts to each jurisdiction’s curriculum, data law, and language requirements.
United Arab Emirates
- Aligns with Dubai Future Foundation’s Human-Machine Collaboration (HMC) classification
- Respects KHDA / ADEK school inspection & AI-use guidance
- Data residency within UAE-hosted infrastructure (PDPL-aligned)
- Arabic + English interface & escalation flows
India
- Maps to CBSE’s K-12 AI curriculum (Grades 9-12) and awareness modules (KG-8)
- Built for DPDP Act 2023 consent & data-minimization requirements
- Configurable for state-board variability (CBSE, ICSE, state boards)
- Multi-language regional rollout via school partner network
Technical architecture
How AIforAll SLM sits inside a school.
Four layers, from the student-facing interface down to the school’s own admin and escalation controls.
Interface Layer
Student & teacher chat/assist UI · age-banded permissions.
AIforAll SLM + Guardrail Engine
Domain-tuned small model · bounded scope · risk scoring in real time.
Data & Security Layer
Jurisdiction-local hosting · encryption at rest & in transit · minimal retention.
School Admin & Escalation Layer
Dashboard for AI Committee · counsellor/parent alert routing · audit trail.
Governance
Who watches the watcher.
ESSS is a tool for human governance, not a replacement for it. A named committee owns every escalation.
School AI Committee
SMT + HODs (or equivalent) review dashboards monthly and own final escalation decisions.
Incident Log
Every flagged interaction — biased output, distress signal, misuse — is logged with resolution status.
Research & Consent Layer
De-identified usage data feeds PhD research under informed consent and institutional ethics review.
Annual Review Cycle
Policy thresholds, tool registry, and escalation workflows reviewed and re-approved each year.
Research & pilot roadmap
From pilot to published framework.
Phase 1
Pilot
1-2 partner schools each in UAE & India; baseline data collection under ethics approval.
Phase 2
Refine
Tune SLM behaviour & ESSS thresholds against pilot findings and stakeholder feedback.
Phase 3
Scale
Extend across the AI Made For All school partner network in both countries.
Phase 4
Publish
PhD thesis findings + open ESSS framework released for wider adoption.
Why AIforAll SLM
Purpose-built beats general-purpose.
| General AI Chatbots | AIforAll SLM + ESSS | |
|---|---|---|
| Scope | Open-ended, unbounded | Bounded to curriculum & age band |
| Visibility | No school-level oversight | Full usage & risk dashboard |
| Wellbeing signals | Not monitored | Proactively scored & escalated |
| Data residency | Often global/opaque | Jurisdiction-local, school-governed |
| Policy fit | Generic terms of use | Mapped to school AI policy directly |
Let’s build a fairer, safer AI classroom — together.
AIforAll SLM and the Early Student Support System are being developed as part of ongoing PhD research under AI Made For All — a community-driven movement bringing ethical, accessible AI education to students, teachers, and parents across the UAE and India.
Arockiaraj Arockiam · PhD Research · AI Made for All · June 2026