Demos

See It Running

Two working systems, not mockups. Each walkthrough is about 90 seconds, runs on synthetic student data, and shows the two things a school should insist on seeing before buying anything: the de-identification layer and the human approval gate, live.

[PLACEHOLDER] Loom recordings are in production. Replace each block below with the Loom embed code. Until then, these placeholders are visible on the live site.
[PLACEHOLDER — Loom embed goes here]

Rubric-to-feedback engine

An AI grading demo for schools: a full class set of essays scored against a teacher-editable rubric, drafted, verified, and held for approval.

What you are watching

  • Twenty essays scored against an editable 4×4 rubric in one batch, in about two minutes
  • Every criterion score backed by a verbatim quote from the student's own writing, verified against the source text before a teacher sees it
  • The edit, approve, or regenerate gate on every single draft, plus a class-level summary and gradebook export

What this proves
Feedback quality does not have to collapse under volume, and a teacher stays the author of record for every comment that goes out.

[PLACEHOLDER — Loom embed goes here]

Bilingual parent-communication generator

A parent communication AI demo: personalized family messages drafted from student context, in two languages, approval-gated before anything sends.

What you are watching

  • Personalized messages drafted for a roster of families, English and Spanish generated side by side
  • Four tone settings and a reading-level control, so the same facts can be delivered the way each situation calls for
  • An approval step before any send, and a delivery log recording what went to whom, in which language, approved by whom

What this proves
Every family can get a specific, readable message in their own language without a staff member drafting each one from nothing.

Both demos run entirely on synthetic student data. No real student record is used in any demonstration.

The architecture under both

Watch where the names go

Both demos are built on the same pipeline we install in schools. The walkthroughs show this boundary on screen, with a before-and-after view of the exact payload the model receives.

How student data is de-identified before any AI call Inside the school's systems, student names are replaced with tokens. Only the de-identified record crosses to the AI model, which returns a draft. Back inside the school's systems, tokens become names again, a teacher reviews and approves, and only then does anything reach a family. INSIDE YOUR SCHOOL’S SYSTEMS Student record Jayden M. De-identify Jayden M. STU-041 names become tokens de-identified only AI model outside your systems sees STU-041 never Jayden M. Returns a draft. Nothing is stored. draft returns Re-identify STU-041 Jayden M. tokens become names Teacher reviews, edits, approves — or not Family receives it No student name ever crosses the dashed line.

The de-identification boundary. Names are tokenized before any request leaves your systems, and re-identified only after the draft is back inside them. The model works entirely on tokens.

The demos show the pattern. The audit finds your workflow.

Two weeks, $2,500, and a ranked roadmap of where automation would return the most hours in your building.

Request an AI Workflow Audit See what it costs