AI
AI assists. Teachers decide.
We use AI where it genuinely saves time — drafting questions, suggesting grades, summarizing trends — and we're specific about where a teacher's review sits in that process, not just that it exists somewhere.
Why we use AI at all
Writing curriculum-aligned questions at the volume a real classroom needs, or drafting a narrative summary across a term's worth of scores and attendance, isn't work a fixed template can do. Where the task genuinely needs judgement-shaped generation, we use AI. Where it doesn't — grading a multiple-choice question, for instance — we use plain, deterministic logic instead, because that's simpler, cheaper, and has no failure mode worth worrying about.
Where AI fits in the workflow
Question generation
Drafts curriculum-aligned questions when the existing bank doesn't already have a good match.
Grading descriptive answers
Suggests a mark and feedback note for short/long-answer responses, checked against a teacher-defined answer key.
Report card narratives
Drafts a summary from a student's real scores, attendance and trend data.
Vidya, the school assistant
Answers plain-language questions about your school's own data.
Reviewed before it's final
Where a teacher reviews before anything is final
Report card and student-analysis narratives are shown to the teacher before a parent ever sees them — this is true today, not a future promise. A teacher can rewrite or discard the AI draft entirely.
Suggested, override available
Where AI suggests, and a teacher stays in control
For question generation and for grading descriptive answers, AI output is available for a teacher to use as-is, edit, or reject — a teacher can rate or flag any generated question, and can review and adjust any AI-suggested mark at any time. We're building toward requiring that review before those items are finalized by default; today, the control exists and a teacher isn't required to exercise it before content is usable. We'd rather say that plainly than imply a stricter gate than what's actually built.
Quality controls in place today
Grounded in your curriculum
Generation prompts instruct the AI to draw only from your school's curriculum source material and not invent facts. That's an instruction we give the model, not a guarantee it's followed perfectly every time — which is exactly why teacher review exists downstream.
Confidence scoring
Every generated question gets a confidence score; low-confidence questions are flagged for review rather than treated the same as everything else.
Deterministic where possible
Multiple-choice and true/false questions are graded by exact match — no AI, no ambiguity, no review needed.
Teacher feedback shapes the bank
A rating or flag from any teacher lowers a question's quality signal for every teacher who might see it next.
Reliable counts, no silent shortfalls
Ask for 10 questions and generation is checked against that exact number before it's returned — if a batch comes up short, one bounded top-up attempt runs automatically instead of quietly handing back fewer than asked for.
Resilient across providers
Question generation and grading route across more than one AI provider, so a single provider's outage doesn't stop the classroom's work.
Vidya: a school assistant that only reads
Vidya answers questions about your own school's data in plain language — attendance patterns, at-risk students, teacher activity. Every one of its tools fetches data; none of them write. It cannot edit a record, send a message, or take an action on your behalf. Requests are also screened for prompt-injection attempts before they're processed.
Data considerations for AI features
Content sent to an AI provider for generation or grading isn't retained by our systems as a prompt log after the request completes. AI-generated content itself (questions, grading feedback, report narratives) is stored the same way any other school record is — scoped to your school, and covered by the same access controls described on our Security page.
FAQ
Honest answers about AI
Can AI grade a student without any human involved?
Objective questions (MCQ, true/false) are graded automatically because there's a single correct answer — that's not really 'AI grading,' it's exact matching. Descriptive answers get an AI-suggested mark that a teacher can review and adjust at any time.
Can AI-generated content be wrong?
Yes — this is true of any generative AI system, and we don't claim otherwise. That's why generation is grounded in curriculum material, confidence-scored, and reviewable by a teacher rather than presented as a final, unquestionable answer.
Does Vidya have access to send messages or change records?
No. Vidya's tools are read-only by architecture — it can retrieve information, not act on it.