Kindred: a privacy-first AI product for schools
Learning plans, drafted by AI, with no child's name anywhere near it. Every sentence cites its report, and the model change halved drafting time.
Inclusion leaders spend hours turning clinical reports into Personalised Learning Plans. Kindred drafts one from about twenty minutes of teacher input.
What it does
- A separate de-identification step strips names and details before anything reaches the model. If it fails, the request stops.
- Every sentence in a plan must cite the report it came from. Uncited sentences are removed before a teacher sees them.
- Moving to a newer model halved the time to draft (156 to 71 seconds) and cut uncited sentences from 8 to 1.
The honest bit
In that same test, the older model invented a whole section with a real-looking page reference. The checker passed it, because it checked that citations exist, not that they support the claim. We wrote that gap up as a high-priority fix ourselves.
Live in an Australian school.