kaedax
← work

An AI-classroom platform that teaches students AI itself — courses, auto-graded quizzes, a breakthrough feed, and a teacher console, live across a school network.

GURUKUL is the Google-Classroom shape rebuilt for one urgent subject: AI literacy. Students were already on ChatGPT; teachers were a year behind; principals had no syllabus to point at. The brief: a structured AI curriculum platform for Classes 8–12 and undergraduate colleges — vernacular-ready, DPDP-respectful with children's data, usable on shared devices and patchy bandwidth, with a teacher dashboard that surfaces who needs help before marks fall. A live demo of the product runs at kaedax.com/demos/gurukul.

stack → Astro + React Postgres + RLS Anthropic API Eval-gated AI tutor CBSE AI 417/843 mapping NEP 2020-aligned DPDP consent ledger Offline-first sync Bhashini-ready i18n
▷ CASE · GURUKUL · BLUEPRINT EDTECH · K-12 + HIGHER-ED
SCHEMATIC An abstract view of the GURUKUL engagement — not a literal product screenshot. Built to communicate engineering shape, not surface design.

outcomes

01

1,284

Active students in pilot month one

02

87%

Average course completion

03

240

Eval cases gating the AI tutor

04

52 min

School onboarding · spreadsheet → live

[ §01 ] the cycle

How 720 hours
actually ran.

  1. Day 01 — 04

    Curriculum architecture + consent model

    scope.agent mapped the AI-literacy syllabus to board structures (CBSE/ICSE/state) with two educators in the loop, and modelled DPDP children's-data consent as first-class database state — parental consent, purpose limitation, no behavioural profiling — before any feature was specced.

    curriculum.spec consent ledger schema 3 courses · 17 lessons
  2. Day 05 — 18

    Student experience build

    build.agent shipped courses, typed lesson content (prose, code labs, callouts, key terms), auto-graded quizzes with instant explanations, streaks and progress, the AI-breakthroughs feed, and the classroom stream. Every lesson written mobile-first for shared-device, evening-study reality.

    41 PRs lesson engine quiz engine + explanations
  3. Day 19 — 25

    Teacher console + AI insights

    The faculty view: live roster progress, quiz analytics, and an AI teaching assistant that flags confusion spikes ('31% of X-A picked the same wrong answer') and inactivity patterns, with suggested interventions. AI tutor responses gated behind an age-appropriateness and syllabus-adherence eval suite.

    faculty console eval suite × 240 cases insight engine
  4. Day 26 — 30

    Pilot rollout

    Pilot across a 14-school network: SSO for institutional accounts, offline lesson packs for low-bandwidth campuses, Hindi UI strings ready behind a flag, and admin onboarding that takes a school from spreadsheet to live classes in under an hour.

    14 schools live offline packs hindi i18n staged

[ §02 ] agent log · selected

What the loop
looked like.

cycle-log · gurukul
archived
T+120h [ OK ] scope.agent curriculum mapped to 3 boards · consent ledger modelled · 17 lessons specced
T+240h [ >> ] build.agent lesson + quiz engine shipped · instant-feedback grading live
T+360h [ OK ] qa.agent tutor eval suite 240/240 · age-appropriateness + syllabus-adherence gated
T+480h [ >> ] build.agent faculty console shipped · confusion-spike insights live
T+600h [ OK ] monitor.agent pilot week 1 · 1,284 active students · offline sync 98.4% reconciled
T+720h [ OK ] deploy.agent 14-school rollout complete · onboarding median 52 min per school

[ §03 ] notes from the cycle

The unusual part of this engagement: the product is publicly demoable. Open the live Gurukul demo → — take a lesson, fail a quiz on purpose to see the explanations, then switch to the teacher view to see what a principal sees.

from the founder

"Our students were already on ChatGPT; our teachers were a year behind. Gurukul gave us a real AI syllabus, guardrails, and a teacher dashboard — live across 14 schools in one term."

— Director · 14-school K-12 network, Kochi