§ For teachers · the other half of the classroom

See the class at concept level.

Qriouso tutors each student individually. Teachers see the whole class in one clear view. Every student × every concept appears on one grid — pseudonymous by default, with names decrypted only in the teacher's browser.

qriouso.school/teacher/analytics

Fig. 01 — Class analytics, captured live: the student × concept mastery matrix for a demo class — nine students, nine physics concepts. Rows are pseudonyms until the teacher's passphrase says otherwise.

subjects mapped

> 20k

concepts

> 40k

prerequisite edges · generated + curated

heatmap view modes

§ 01 · See the class

Every student, every concept, one legible grid.

One matrix, four readings. Flip the same class between mastery, status, quiz success, and spaced-repetition health — the rows never change, only the question you are asking.

view modes — mastery status quiz success spaced repetition

Concept coverage panel listing problem areas where the class is struggling, with per-concept success rates between 38 and 62 percent on German-language physics concepts. — Fig. 02 — Concept coverage for the same class: problem areas surface with per-concept success rates, so the next lesson starts where the class actually struggles. Austrian curriculum — German-language content, multilingual by design.

the name column

Student names stay in the teacher's browser.

Qriouso analyses pseudonymous student IDs. Only the teacher can unlock the name column with a passphrase, and decryption happens locally in the browser. The server stores encrypted names and never receives the passphrase.

pseudonymous student ID encrypted name unlocked in the teacher's browser · unreadable to the server

The server never sees a name.

§ 02 · Assign the path

Assignments that follow the graph.

Behind every assignment sits the prerequisite graph — generated deterministically, then teacher-curated. You pick the goal; the wizard brings what the goal depends on.

§ 02 · the wizard

Pick the goal, inherit the path.

The assignment wizard distinguishes hard prerequisites, which gate progress, from soft ones, which inform a readiness ratio. Assign the destination and the route comes with it — no hand-built sequences.

Assign-learning-path wizard on step two of four, with 70 concepts selected in prerequisite order and a note that concepts build on each other. — Fig. 03 — The assignment wizard at step 02 of 4 (subcompetency → concepts → schedule → quiz review): 70 concepts selected, held in prerequisite order — concepts build on each other, so the sequence is locked.

§ 02 · the path

Struggle routes backwards, by design.

Each node carries a state, and a struggling student is routed back to prerequisite concepts automatically — remediation is the default behaviour of the path, not a referral the teacher has to remember to make.

node states — mastered in progress ready next blocked

Teacher curriculum prerequisite graph filtered to Physics through grade 8, showing 57 competencies, 51 dependencies, and 1,145 linked concepts across grades 6, 7, and 8. — Fig. 04 — The live Austrian curriculum graph, filtered to Physics through grade 8: 57 competencies, 51 dependencies, and 1,145 linked concepts. Teachers can select any node as a target and inspect every prerequisite path.

§ 03 · Build the material

Build the material, keep the craft.

Quizzes, worksheets, prompts, media — authored at one bench, and every result flows back into the same graph the class runs on.

§ 03 · a

Twelve question types, one flaw detector.

From multiple choice to image hotspot and label-image with AI-generated images, LaTeX throughout. Item-writing-flaw detection flags weak items before students ever see them, and generation runs async so a full quiz never blocks the tab.

Quiz-builder question editor with an AI-generated animal-cell diagram and a rectangle drawn over the nucleus as the correct hotspot region. — Fig. 05 — A hotspot question in the quiz builder: an AI-generated animal-cell diagram, the correct region drawn straight onto the image.

§ 03 · b

One source, differentiated output.

Worksheets come off the same bench: set the topic once, generate three variants, and assign the right one per student — difficulty groups included.

Differentiated-worksheets page with a configure-preview-assign stepper and settings for subject, topic, grade level, differentiation method, and difficulty groups. — Fig. 06 — Differentiated worksheets: configure → preview → assign. Three variants from one topic, matched to difficulty groups.

§ 03 · c

The workshop drawer.

A prompt library for the phrasing you refine over years, workflow automation for the prep that repeats, and an AI media library for images and figures — one drawer, open beside whatever you are building.

Prompt library with tabs for my templates, shared with me, and school library, plus a safety banner stating that students never receive prompt templates and assignments stay blocked server-side until the guard check and teacher approval pass. — Fig. 07 — The prompt library: templates shared across a school — and a hard rule in the banner: students never receive prompt templates; assignments stay blocked server-side until the guard check and teacher approval are proven.

§ 03 · d

Results flow back.

Passing a concept quiz lights the corresponding heatmap cell, and mastery feeds the spaced-repetition review queue. Assessment is not a separate ledger — it is the same graph, updated. The heatmap in Fig. 01 is where those results land.

§ 04 · A chat of your own

The same tutor, on your side of the desk.

Teachers get the tutor as a GDPR-first chat workspace of their own — the same personas, with adaptable tone and depth: easier or harder language, plainer or more advanced vocabulary.
The live demo lets you drive a demo student against the live tutor — every persona, every guardrail — and watch each prompt run the non-spoiling pipeline, stage by stage, before you assign anything to a class.
Oversight is structural: tutoring conversations are inspectable by design, not by exception.

Non-spoiling trace panel: a prompt demanding the complete exam answer runs through the pipeline stages from received to decision, ending in a green safe-reply verdict because the tutor refused and asked a counter-question. — Fig. 08 — A real run, captured live: the demo prompt demands the complete exam answer; the trace shows every pipeline stage — received, prerequisites, generation, spoiler check, leak check, decision — and the reply that ships is a refusal with a counter-question.

Student home screen with eight tutor-style chips from Socratic to Practical, a dual-engine toggle, and a chat input asking what the student would like to learn. — Fig. 09 — What the class sees: all eight tutor styles on the student home — tone changes, the guardrails never do.

§ 05 · Data stewardship, by construction

Stewardship as architecture.

A classroom platform holds minors' learning data. The protections here are structural — built into how data flows, not appended to a policy page.

Minimization

Pseudonymous by default.

Analytics run on pseudonyms; real names exist client-side only. Logging is one wide-event line per request — no ad-hoc logging of student text scattered through the codebase.

Provider posture

Zero data retention.

LLM calls run under zero-data-retention agreements (AVV/DPA): prompts are not stored by providers and never train models.

Control

DSGVO/GDPR tooling built in.

Data export, erasure routines, and retention jobs ship with the platform — the rights a school must honour have handlers, not tickets.

Posture and mechanisms — not a checkbox.

§ 06 · Get in touch

Pilot it with your class.

One class, one subject, one term. Austrian curriculum underneath, pseudonymous analytics by default, and a tutor you can interrogate before your students ever meet it.

Start a pilot conversation How tutor models qualify →

steven.ludwig.school@gmail.com