What our AI does — and what it doesn't touch
District technology directors ask exactly these questions, so here are direct answers: which AI we use, whether student data is sent to it, and whether student work trains the models.
The short version
Which AI vendor?
AI features are powered by leading third-party language models (including providers such as OpenAI, Google, and Anthropic) through our platform's integration layer. We select the model per feature and can tell you which provider powers a given tool on request.
Is student data sent to it?
For most practice, no — auto-grading and the arcade run on the app's own logic. Socratic hints sends only the teacher-authored question text (never the student's typed answer), so no student data reaches a model, and a teacher must enable it per student. Handwritten-work analysis is the one place student work reaches a model — it's teacher-initiated (the student uploads a photo, and the AI runs only when the teacher clicks Analyze), always on a no-training provider tier, and student names are never included in what is sent. A district with its own reader has photos sent only to that reader; if it is unavailable the work is queued for it, not sent elsewhere.
Is it on by default?
No. AI generators are optional, premium tools a teacher chooses to use. Core teaching and learning — assigning, practicing, auto-grading, and reporting — works without any AI at all.
Does it train on student work?
No. We do not use student work to train our own models. The one feature that sends student work (handwritten-work analysis) runs only on a no-training provider tier — one whose terms do not allow training on customer-submitted data — and never on a free model tier. Socratic hints and the teacher-only generators run on free-tier Google Gemini, which may allow product improvement under its terms, so student data is never sent there (Socratic hints sends only the teacher-authored question, never the student's answer). We're happy to share the relevant provider terms during your review.
What each AI feature sends & keeps
Most features are teacher tools that never see student data. Socratic hints sends only the teacher-authored question (not the student's answer) on the free tier. The one feature that sends student work — handwritten-work analysis — is teacher-initiated and runs on a no-training provider tier. Here's exactly what is sent and what is stored.
| AI feature | Who | Sent to the model | Stored |
|---|---|---|---|
| Warm-up, quiz & worksheet generators | Teacher | Grade, topic, standard code, and the teacher's prompt. No student data. | Generated questions and any AI-made illustrations. |
| Handwritten warm-up & worksheet converters | Teacher | A photo or file the teacher uploaded, plus the teacher's own handwriting legend. No student data. | A draft assignment the teacher reviews before publishing. |
| Answer-key generator | Teacher | The teacher's questions to produce an answer key. | The answer key, stored server-side for the teacher only. |
| Intervention plan generator | Teacher | Aggregated, de-identified class gaps by standard — not student names. | A suggested reteach plan for the teacher. |
| Assessment score import & intervention groups | Teacher | Nothing. The assessment export (Star, CMAS, Illuminate, i-Ready, MAP, or another state test) is parsed in the teacher's browser — no names, emails, IDs, or scores are sent to any AI service. Grouping suggestions are computed without AI. If a teacher creates a draft reteach lesson from a group, only the grade, skill description, the group's anonymized assessment levels, and group size are sent — never names, emails, or IDs. | Assessment results on our servers, visible only to the importing teacher, their co-teachers, and site admins. Students and parents cannot read them, and the teacher can delete them at any time. |
| Relay question generator | Teacher | Grade and learning target code. | Race questions for the live game. |
| Handwritten-work analysis | Teacher | A photo of a student's handwritten work and that student's handwriting legend. The AI runs only when the teacher clicks Analyze — the student's upload itself never calls the model. Always a no-training provider tier; student work is never sent to a free model tier. No student name or identifier is sent. A district with its own reader has photos sent only to that reader (signed, no identifiers); if it is down the work is queued rather than sent elsewhere. | The analysis and feedback on the student's submission record, visible to the student and their teacher. |
| Socratic hints | Student | Only the teacher-authored question text — never the student's typed answer — so the model can give a hint without revealing the answer. A teacher must enable the Socratic tutor for a student before it's available. | The hint is shown to the student; it is not stored long-term. |
"Teacher" tools and Socratic hints run on free-tier Google Gemini, which may allow product improvement under its terms — so no student data is sent there (Socratic hints sends only the teacher-authored question, never the student's answer). Handwritten-work analysis, the one feature that sends student work, runs only on the credit-based no-training tier (Anthropic / OpenAI), whose terms do not train on customer-submitted data. The answer-key generator runs on that same no-training tier.
Teacher handwriting legends (the digits, letters, and math symbols a teacher writes once so the converter can read their handwriting) are never stored on our servers — the photos go straight to the AI check and stay only in the teacher's own browser. Student handwriting legends are stored encrypted — the stored file is unreadable ciphertext without a decryption key that only the student and site admins can access — and student work photos stay private. Both are read only when a teacher clicks Analyze. Teacher-uploaded question images use the same encrypted storage; the decryption key reaches only the teacher, site admins, and students assigned that warm-up.
A teacher reviews everything
- Nothing AI-generated reaches students until a teacher reviews and publishes it.
- Teachers can edit or delete any AI-generated question before it's assigned.
- Auto-grading (multiple choice, numeric, matching, graphing) runs on the app's own logic — it does not call an AI model.
- Handwritten-work analysis is the one place student work reaches a model — and it's teacher-initiated, with no student names sent. It always runs on a no-training provider tier, never a free model tier, and only when the teacher clicks Analyze. Districts with their own reader have photos sent only to that reader.
- The student 'My Math Journey' arcade mode is procedurally generated in code and uses no AI at all.