Privacy & de-identification
The privacy comes first. Then the AI.
One de-identification layer sits under every Kindred tool. No identifying student detail reaches the model, on any product, whether you're planning a lesson, marking a class or drafting a learning plan.
The privacy spine
No identifying student data reaches the AI.
Right now, the shortcut is pasting student work and reports straight into ChatGPT. That's the status quo Kindred replaces, and it's a different category of risk. One de-identification layer sits under every Kindred tool, so the privacy story is the same whether you're planning a lesson or marking a class.
No identifying detail reaches the AI
Names, year level, school, dates: the things that make a student identifiable are removed at the edge before any AI processing. The model never sees the student.
You approve what gets removed
For marking, Kindred shows you every identifier it's about to strip. You confirm it before a single word goes to the model. It's your attestation, in your hands.
The teacher is always the decision-maker
Kindred drafts and scores a first pass. It never finalises a lesson or a mark. Nothing leaves as final until a person has reviewed and approved it.
It says what it can't do
Rather than fabricate, Kindred declares the criteria it can't judge or the claim it can't support, and routes it back to you. Nothing unsubstantiated survives.
Nothing is quietly kept
Marking runs are session-only: student work, scores and the re-identification map are gone when you close it. The only lasting artifact is the file you export.
Australian, and built for the rules here
Made in Melbourne against the Victorian DET position on AI, the ST4S AI module, the NCCD, and the National AI in Schools Framework.
How it works
The same four moves, whichever tool you reach for.
Every Kindred tool works the same way. The AI drafts. Identifying detail never reaches it. A person approves.
Bring the real material
The curriculum outcome you're teaching, or the rubric and the class set of work. Kindred reads PDFs, DOCX, slides and scans, whatever you already have.
Identifying details come off first
Before anything reaches the AI, student names and identifying details are stripped at the edge. For marking, you approve exactly what gets removed. The model never sees who the student is.
Kindred drafts the work
The lesson, deck and worksheet, or the scored marksheet with a comment per student. It shows its working, and it says plainly what it can't do rather than making something up.
You review, edit and own it
Nothing is final until you say so. Edit it into your voice, override any call, export what you need. You are always the teacher of record; Kindred just did the first draft.
FAQ
Privacy, answered
- What does 'de-identification at the edge' mean?
- Identifiers (names, year level, school, dates) are detected and removed before any data leaves for AI processing. A deterministic pass seeded with known names runs first, then a medical-grade model catches the rest. The AI only ever sees de-identified text.
- Isn't pasting into ChatGPT the same thing?
- No. Pasting a student's work or report into a general chatbot sends identifiable information to a third party. That's the status quo Kindred replaces, and it's a different category of risk. Kindred strips identifiers first, by construction.
- Does Kindred keep student data?
- The marking tool is session-only: student work, scores and the re-identification map are gone when the session ends. The lasting artifact is the file the teacher exports. Usage counters that persist are PII-free.
- Is Kindred aligned to Australian rules?
- It's built in Melbourne against the Victorian DET position on AI, the ST4S AI module, the NCCD and the National AI in Schools Framework.
A privacy story your DPO can actually sign off.
Want the detail: the de-identification pipeline, the compliance posture, a DPIA? Get in touch and I'll walk your school through it.