Observe
Skill, recent attempts, hints, confidence and previous support.
How SafeSpark™ decides
Every recommendation should be understandable.
SafeSpark™ uses approved activities, a visible learner-state record and parent controls. It does not hide routine choices behind a vague “AI-powered” label.
The decision sequence
Observe → interpret → support → check → update → plan.
SafeSpark™ gives the smallest useful amount of help and reduces it after secure responses.
Interpret
Decide whether to continue, simplify or review a prerequisite.
Support
Choose smaller steps, objects, visuals or read-aloud help.
Check
Record the answer, confidence and whether the support helped.
Update
Adjust the individual skill state and support preference.
Plan
Select the next approved activity or stop before overload.
Learner-state model
Skills are tracked separately.
A learner can be secure in one area and still need support in another. The state is a planning signal, not a diagnosis.
New
Attempted
Developing
Nearly secure
Mastered
Needs review
Live Decision Inspector
Inspect the recommendation using the local learner state.
No profile yet? Load the seeded Maya demonstration.
SafeSpark™ may
- Change explanation style and difficulty
- Offer approved hints and read-aloud support
- Revisit prerequisites and plan review activities
- Explain the recommendation to a parent
SafeSpark™ may not
- Browse the open internet or run unrestricted child chat
- Diagnose a condition
- Contact strangers or share data without authorised action
- Replace teachers or make high-stakes placement decisions
Current working engine
Approved activities, explicit mastery states, deterministic planning, confidence signals, support-route memory and parent controls.
Human control
Parents can prefer or disable a support route, override the next skill, reset an inaccurate state, export the record or delete it.
Open parent controls →