1. AI features
| Feature | Information used | Purpose and limit |
|---|---|---|
| Student doubt support | Student questions and up to five completed exchanges in the same thread | Study assistance; not a guaranteed correct answer |
| Subjective and IELTS Writing evaluation | Student response, question, answer guidance, rubric, and assessment context | Teacher-reviewable scoring and feedback |
| Test and content creation | Teacher instructions, course material, questions, and generation constraints | Draft content requiring teacher review |
| Progress and intervention support | Grades, completion, attendance, and selected learning signals | Suggestions for an authorized teacher or administrator |
| OCR and media generation | Authorized answer images, listening scripts, or visual instructions | Transcription, extraction, or generated audio/media |
2. Providers
Depending on deployment configuration, AI processing may use GitHub Models, Groq, OpenAI, Gemini, or Wit.ai. Providers can change as models, reliability, contracts, and safety requirements change. The service operator must review provider retention and model-training settings before commercial use.
3. Human review
- Teachers can review and override AI-assisted grades and feedback.
- Generated assessments and learning material should be reviewed before publication.
- Risk scores and interventions are support signals, not automatic disciplinary decisions.
- AI doubt responses should be checked against trusted educational material.
- AI must not make final safeguarding, medical, legal, admissions, or employment decisions.
4. Data minimization and logging
AI routes are intended to send only the content required for the requested task. Operational logs are designed to record provider, model, timing, failure category, and opaque actor references without recording raw prompts, student answers, essays, chats, or credentials.
5. Known limitations
AI can hallucinate facts, misunderstand handwriting, miss cultural or language context, apply rubrics inconsistently, and produce repetitive or biased suggestions. Confidence indicators are not guarantees. Users should report consequential errors and request human review.
6. Questions, correction, and objection
Use the Data Rights page to ask what AI-related personal data is processed, request correction of underlying records, raise a concern, or request eligible consent withdrawal.