Skip to content

Case study · school AI product

Bose

A school AI product built to shorten the learning loop and make assessment easier to trust. Bose teaches from school material, turns weak chapters into practice, calculates marks from stored rules and keeps the record behind every result.

school AI product teaching + assessment demo-ready private repo

Demo-ready

Student, teacher, admin and parent journeys built

Adaptive

Mastery and recent work shape the next explanation

Traceable

Answers, rubrics and calculated marks stay connected

Private

Real names stay in an encrypted local vault

why I built Bose.

I built Bose after noticing two gaps in the way schools use technology.

01 · trust

Students should be able to understand a result.

The student protests around exam transparency made the problem impossible to ignore. A score should not feel like the end of a closed process. The answer, marking rule, calculated score and recorded exam events should form a trail that a teacher can inspect.

02 · learning

Feedback should arrive while it can still help.

Students often wait for the next class or marked paper to learn what they missed. Bose uses recent work and chapter mastery to choose the next explanation, hint or practice set, so the learning loop keeps moving.

The product idea: connect teaching, practice, exams and marking in one system. Students get help based on what they actually know. Teachers get the evidence behind every result.

how the learning loop works.

One learning loop, not a stack of disconnected tools. Bose carries context from teaching into practice, from practice into assessment, and from each result into the next lesson.

01

protect identity

Real names stay in an encrypted vault. The rest of Bose uses a private student ID.

02

understand the learner

School material, recent work and chapter mastery shape the response.

03

choose the next step

Bose can explain, hint, create practice, build a paper or run an exam.

04

keep the result clear

Answers, scoring rules, marks and exam events stay together for review.

how marking works.

Every result begins with evidence. Bose checks the submitted answer against the stored answer key or rubric, applies the assessment rules and saves how the mark was reached.

01 · INPUT

recorded answer

The student's submitted answer becomes the source data.

02 · RULES

answer key or rubric

The assessment's stored scoring rules define what earns marks.

03 · RESULT

calculated mark

The same rules are applied and the result is saved with its inputs.

04 · REVIEW

teacher checks

The teacher can inspect the answer, score and recorded exam events.

Why this is more trustworthy: every mark has a visible path back to the recorded answer and scoring rule. Teachers can review the result instead of trusting a black-box score.

privacy engineering.

This is Bose's main privacy rule: a student's name never reaches an external model.

ON DEVICE / SCHOOL STUDENT INPUT question + real identity AIRLOCK AES-256-GCM VAULT REAL ID STOPS HERE REAL IDENTITY SCRUB strip before persist RIGHT TO FORGET RE-IDENTIFY ANSWER BACK PSEUDONYMOUS ID TRUST BOUNDARY OFF DEVICE EXTERNAL MODEL scrubbed question + ID ANSWER UP CROSSES: scrubbed question + pseudo ID NEVER CROSSES: name, class, identifiers

The privacy boundary. A real name stays in the encrypted vault. Only a private student ID and scrubbed question can reach an external model.

The rule is enforced in four places:

  • The vaultReal identities stay in one AES-256-GCM encrypted store. The rest of Bose uses a pseudonymous ID.
  • Scrub before savingBose removes identifying details before saving data or calling an external model.
  • On-device camera checksThe browser records integrity signals. Video never leaves the device.
  • Right to forgetIdentity lives in one vault, so deletion is direct and complete.

No school, student data or private repository detail is shown here.

what's live.

The demo covers the full school loop, from a student's first question to a teacher reviewing the result:

privacy boundary

identity airlock

Keeps real identities in an encrypted vault and removes personal details before any model call.

school material

answers from school material

Uses the class syllabus and answer bank, so explanations stay tied to what is being taught.

graduated help

adaptive tutor

Uses recent work and mastery to decide whether to guide, hint or explain the answer in full.

paper generation

question-paper engine

Reads the shape of past papers, then creates new papers with answers and marking rubrics.

assessment + marking

online tests and marks

Checks recorded answers against the assessment rules, calculates marks and keeps each result reviewable.

browser-only

exam mode + proctoring

Runs timed, randomized exams and records tab switches, fullscreen exits, copy-paste, device changes and camera-based integrity signals.

class progress

visible learning progress

Turns preparation, learning and assessment activity into points, mastery and a class scoreboard.

approval gate

teacher automation

Builds worksheets, suggests grades and groups common doubts while teachers review student-facing decisions.

the bigger goal.

Bose is demo-ready today. The next layer is a school-wide view of how students are learning with AI. Preparation, knowledge gained and assessment activity will contribute to a student score. The average will show a school's progress over time and allow comparison with schools in the same area.

This layer is planned. Teachers will be able to inspect the activity behind each score, so the number remains useful instead of becoming another unexplained rank.