← Projects

QBO-A04 · Quobo Labs fleet

Quobo Labs Education

Labs applicationField study

A private AI workspace for special-ed paperwork: every draft is grounded in the student's own record, and nothing is written without the teacher's one-click approval.

Out in the field being validated — the story is below.

Chapter one

The problem

Special-education teachers run two jobs. The first is teaching. The second is a paperwork practice a small law firm would recognize: IEPs, progress reports, behavior assessments, intervention plans, meeting notes, parent letters — per student, per year, on legally binding deadlines.

The moment it hurts is specific. It's 8pm on a school night, the annual review is Thursday, and the teacher is rebuilding one child's story from a filing cabinet, last year's PDF, and memory — then writing prose about it for hours, alone. The document has to be individualized, defensible, and consistent with everything ever written about that child. Nothing in the building helps with the writing; the district systems only care that the finished form gets filed.

And the obvious shortcut is off the table. None of this can be pasted into a chatbot, because it's protected student data. The one tool that could help with the writing is the one tool the writing isn't allowed to touch.

The mechanics

How it works

  1. Three screens, no chat homepage

    The whole app is a roster, a file tree, and a settings page. There is no prompt box to stare at. The AI works under the hood — drafting, editing, and summarizing as a side effect of normal document work — so the teacher's day starts with students, not with prompt engineering.

  2. Each student's record is a living corpus

    Every document — the IEP, the behavior plan, last quarter's progress report, the notes from October's meeting — lives as one body of writing per child. It's the teacher's filing system and the AI's memory at once: everything is indexed so that whatever gets drafted next is grounded in what is actually true about that student.

    The corpus: one student's file tree, ten document types

  3. The AI drafts whole documents from the child's history

    Ask for a draft and the AI assembles what it knows — the student's profile, guidance for how to write for their needs, and the most relevant passages from their record — and proposes a complete document, or fills a single empty section in place. Drafts aren't generic; the new IEP agrees with the behavior assessment because they share a memory. It drafts documentation; it never diagnoses.

  4. Nothing is written without approval

    Every AI draft is a proposal until the teacher clicks approve. For small changes, select any passage, say what should be different, and review the exact edit as a diff before applying it. Every change — human or machine — lands in an append-only version history, so there's always a record of what was proposed and what the teacher chose.

    A surgical edit: selected text, instruction, diff, approve

  5. Ask questions across the whole caseload

    A quiet assistant can answer questions over everything the teacher has written — "what did we change after the October meeting, and did it work?" — with citations that jump to the exact cited paragraph. The caseload's institutional memory answers in seconds instead of an afternoon of re-reading.

  6. Privacy is the architecture, not a checkbox

    Names, ID numbers, and birthdates are redacted before any text leaves for the AI. Students can be displayed as initials only. Every table is locked to the one teacher who owns it, and an audit log that contains no student data records every action. Using AI on student records without handing student records to an AI is the whole design.

Field notes

Would this fit your life?

A middle-school SPED teacher, caseload of fourteen

The moment
A student's annual IEP review is in three days, and the current IEP is a year old.
What they do
She opens the student's record, gets a full draft grounded in everything she wrote all year, then walks it like an editor — tightening weak paragraphs with small, diff-previewed edits and approving each one.
The payoff
The blank-page night becomes an editing hour, and the draft already agrees with the behavior plan and last quarter's report because they share a corpus.

An elementary teacher facing an escalating behavior

The moment
Repeated incidents mean a student needs a formal behavior assessment, and then a plan built on it.
What they do
He drafts the assessment from the student's profile and every meeting note that mentions the behavior, refines and approves it — then drafts the intervention plan the same way, grounded in the assessment he just finished.
The payoff
The plan actually follows from the assessment, with a version trail showing exactly what was written when — consistency that survives scrutiny.

A privacy-anxious teacher — or her district's tech director

The moment
"You're putting IEPs into AI?"
What they do
She walks through the settings: initials-only display, redaction of names and IDs before anything reaches the model, a per-teacher locked database, an audit log with no student data in it, and an approval click in front of every write.
The payoff
A straight-faced answer to the scariest question in the room. The privacy architecture is itself a use case.

No varnish

From the lab notebook

Where it stands

As of a full codebase audit in July 2026: a finished, polished v1. All planned phases complete, a real test suite — unit, end-to-end, mobile viewport, accessibility, and a negative test proving one teacher can never read another's data — plus a deploy checklist and a written privacy posture. It was built for exactly one user, and it isn't publicly available; there's nothing to sign up for.

The field study

The current work isn't code — it's a discovery study with a practicing special-education teacher, designed to find out where the build's assumptions are wrong without leading her. v1 bets that writing the documents is the bottleneck. The study tests whether the real pain is the writing, or the deadlines, the duplicate data entry, or the gathering of scores and inputs — each of which points at a different product.

What it doesn't do yet

It drafts beautifully, but it doesn't talk to the district and state systems where the paperwork legally lives — a finished draft still gets re-typed into the official forms by hand. There's no way to upload the thick folder a transfer student arrives with, progress lives as prose rather than numbers, and it assumes one student, one document, one teacher at a time. Whether those gaps are dealbreakers or footnotes is exactly what the field study is for.

Next in the fleet

QBO-P01

Events by Le Voyage

Live

A real events business running on lab-built code: a Louisiana dance-cruise company whose whole back office — bookings, payments, group finance, an AI assistant — is the lab's investment in the business it operates.

Continue the tour →