Problem
The platform wanted AI-powered help for learners, but its users include children. Every answer had to be appropriate, personal data had to stay protected, and the feature had to behave predictably.
Solution
- AI agents with strict input and output guardrails and age-appropriate response rules.
- Data minimisation: models receive only the context a task needs.
- A provider-agnostic design with fallbacks for uninterrupted service.
- Logging and review tools so the team can audit how the assistant behaves.
Result
The platform gained AI features designed around the needs of a sensitive user group, with safeguards the team can inspect and adjust.