How are you measuring the progress of your students?
Infrastructure for next-generation education.
Build, run, test, fix — with the record of learning.
Free to begin · no setup · nothing to install
Confirmed work
Counts a trainer approved — never an estimate
318resolved & confirmed
47sent back to re-learn
6m 20smedian stuck → resolved
9waiting on a trainer
resolved on their ownresolved after a hint
Buildlearn by doing
Runapply it
Testsee what broke
Fixand keep the attempt
Every attempt keptThe failures included, not just the one that worked.
Confirmed by a personA trainer in the room, never the model.
Record of learningThe road to the answer, not just the answer.
Photo · code · one record
Make progress visible.
Learners build, run and fix their work inside Kovaan OS. Teachers see the road to the answer, not just the answer.
Get startedFree to begin · no setup · nothing to install
What a camera seesWhat the schema reads
The portals
Three doors into one system.
Same platform, same records, three ways in. When a trainer confirms a fix, all three move at once — nobody re-enters anything, and nobody waits until the end of term to find out how it went.
01 · School & programme
Approve people, watch the programme.
Principals, coordinators, EdTech ops
- Approve trainers, create batches by grade and section
- Lab utilisation and confirmed progress per batch
- Parent reports, built from confirmed results only
02 · Trainer workspace
Prep the class, confirm the work.
EdTech-employed and in-house trainers
- A prep sheet before the session — faults, demos, checklist
- Approve their own students into the batch they teach
- One screen, one tap to confirm or send back a fix
03 · Student portfolio
Get unstuck, keep the proof.
Students · parents read-only
- Photograph the bench, get a question rather than an answer
- The badge appears immediately, provisionally
- Every attempt stays — the failures included
One confirmation, three portals. A trainer taps approve; the badge turns confirmed, the school’s counts move, and the resolution enters the dataset. There is no nightly job and no second entry.
Descriptions, not screenshots. Each door lists what that portal does and opens its sign-up; none of the three is rendered here, and none of this band reads a school’s records.
Inside the school portal
Goals on one side. Confirmed work on the other.
A principal sets what the term is meant to cover. Kovaan reports against it from events that already happened — every figure beside this is a count of resolutions a trainer confirmed, not an estimate of what a learner can do.
- One confirmation, three portals, nothing re-entered by hand
- Framed to improve the programme, never to grade the trainer
- Parent reports built from confirmed results only
School · Vidya Vihar · Term 26 batches · 142 students
Faults resolved and confirmedtrainer-verified · enters the dataset318
Sent back as a re-learn taskrejected · not counted as complete4712.9%
Median time from stuck to resolvedhint shown → student marked resolved6m 20s
Waiting on a trainer right nowqueue depth across 6 batches9
Where rejections cluster
M3 · Series circuits18 of 28
M5 · Serial output11 of 27
M2 · First sketch4 of 26
Modules needing curriculum attentionnever trainer accuracy
Illustrative figures. This panel is drawn for the page, not read from a school. The layout, the wording and what each figure counts are the product’s; the numbers are invented, and no tenant’s records are queried by a public page.
Three ways to see
A build can break in three places.
The wiring can be wrong, the sketch can be wrong, or both look right and the board still misbehaves. Each needs a different kind of evidence — and only one of the three can confirm itself without a human.
Photo · /api/diagnoseTrainer review
POST /api/diagnose inputType: photo faultDetected true faultId missing_resistor confidence 0.96 ← carries no signal verification pending
What the student sees“Trace the path from pin 13 to your LED. Is anything sitting between them?” — the hint first, never the fix. The answer unlocks after sixty seconds.
Who confirms a fix is decided by which input produced it — never by how confident the model claimed to be.
Illustrative output. The three traces are written for this page — the field names, the gate and the routing are the product’s, the values are invented, and nothing here calls the API.
The photo input, working
It lights up. That is the problem.
A student can make an LED light by bypassing the resistor. It works. Confirming that as correct would teach the system to certify bad engineering — which is why the photo input is the one that still goes to a person.
- Graded against this kit's own fault library, not against what a model happens to know
- A question first — the fix unlocks after sixty seconds, or the moment they ask
- Confidence is reported and decides nothing about who reviews it
Workspace · Assignment 2 — LED circuitNot yet diagnosed
Before anything is sent
This is what a student photographs at the bench. Diagnose it and the answer is matched against this kit’s own fault library — never against what a model happens to know.
An illustrative diagnosis. The photograph, the reading and the fault id are drawn for this page. The flow is the product’s — a question before a fix, a confidence that decides nothing, a badge that stays provisional until a trainer confirms it — but nothing here calls the API or enters a dataset.
Where a build fails
What we can see, and what we cannot.
A cell is filled only once that input has been shown to tell the fault apart from its absence. Most of this map is empty on purpose.
Where it fails
Photo
Code
USB / Bluetooth
In the softwareit renders wrong, or runs and does the wrong thing
not covered
built, dark
not covered
On the boardit will not upload, or uploads and misbehaves
not covered
built, dark
v2
A part missing from the paththe resistor that was never fitted
shown detectable
not covered
not covered
A part the wrong way roundthe LED reversed in its holes
ruled out
not covered
not covered
shown detectable — livenext input we buildbuilt, waiting on a measurementtried and removed
One of the three confirms itself. A device’s own report is deterministic text off the board — it cannot hallucinate, so those results need no human at all. That is why USB and Bluetooth are next, and why every result today still goes to a teacher.
For the learner
Write it, run it, fix it — all in one place.
No local setup, no toolchain, no “it works on my machine”. The editor, the compiler and the guidance are the same screen. When something fails, Kovaan points at the line and asks a question rather than handing over the answer.
- Assignments arrive in the workspace; nothing to download
- A compile error is a fact — the rest still goes to a teacher
- Getting it wrong four times is kept as a record of effort, not a penalty
Workspace · Assignment 411 min · 4 attempts · 1 hint
11:04Did not compilemissing ';' — line 8error
11:07Compiled, checks failedamber declared on line 3, never written to2 of 4
11:12Compiled, checks failedamber high for 0 ms — one hint takenhint used
11:15Submittedall 4 checks — awaiting the teacher's confirmationprovisional
An illustrative session. The four attempts, their timings and the hint are drawn for this page. What the panel shows — every attempt kept, a hint counted rather than hidden, and a submission that stays provisional until a teacher confirms it — is the product’s.
Attempt history
The road to the answer, not just the answer.
Every submission is stored against the problem it belongs to. A teacher opens one student’s work and reads the sequence — what broke, what they tried next, how long it took.
Kovaan OS · Workspace · Assignment 4 — Traffic light controller
4 attempts · Grade 7 student
1// traffic light controller
2const int RED = 11;
3const int AMBER = 12;
4const int GREEN = 13;
5
6void loop() {
7 digitalWrite(RED, HIGH);
8 delay(4000)
9 digitalWrite(RED, LOW);
10 digitalWrite(GREEN, HIGH);
11}
Run output
error: expected ';' before 'digitalWrite'
Kovaan asks
Line 8 ends one way and line 9 ends another. What does the compiler need at the end of a statement?
Show me the fix · unlocks in 41s
Photo — liveCode paste — in developmentUSB serial — in developmentevery attempt kept · 11:04
An illustrative attempt history. The four submissions, their timings and the questions are drawn for this page. What the window shows — every attempt stored against its problem, a question before a fix, and a badge that stays provisional — is the product’s.
The completion journey
Most platforms keep the answer. Kovaan keeps the road to it.
The same four attempts a teacher reviews become the thing the child keeps — an account of what they tried, what broke, and the moment it worked.
not running
Four attempts, eleven minutes.
Every one of them kept, including the three that did not work. Turn the page to read how it went.
01 / 05
An illustrative journey. The four attempts, their timings and the badge are drawn for this page. What the book shows — every attempt kept, the failures included, and a badge that stays provisional until a teacher confirms it — is the product’s.
Teacher workspace · Learner 049 assignments this term
Statement terminator missed7×
across 5 assignmentsstill happening
Loop bounds off by one4×
across 3 assignmentsimproving
Variable used before assignment2×
across 2 assignmentsresolved
Where ten minutes would go furthestTerminators are costing this learner about nine minutes a session, and it is the only pattern not improving. Everything else is on its way down.
An illustrative learner. The patterns, the counts and the note are drawn for this page. What the panel reads across — every attempt, not one submission — is the product’s; the learner is not.
For the teacher
The mistake they keep making, named.
One wrong answer tells you very little. The same wrong answer seven times across five assignments tells you exactly where to spend ten minutes. Kovaan reads across every attempt a learner has made and surfaces the pattern — so instead of marking work, a teacher gets a place to start.
- Pat across assignments, not just within one submission
- Shows whether a weak point is still open, improving, or gone
- Works at class level too — if half the room repeats it, it is the teaching
Back to the learner
They can watch themselves getting better.
Because every attempt is kept, improvement stops being a feeling and becomes a shape. Learners see how much they resolved, and how much of it they resolved without asking — which is the number that matters, and the one that should go up.
- Measured visible during the term, not summarised after it
- Independence is what is measured, so the goal is needing Kovaan less
- The record belongs to them and follows them out of the institution
Learner · Your progresslast 6 weeks
W1W2W3W4W5W6
resolved on their ownresolved after a hint
68% of week six resolved on their own, against 16% in week one.
An illustrative learner. The six weeks are drawn for this page. What the chart measures — attempts resolved on their own, kept across the term rather than summarised after it — is the product’s; this learner is not.
The platform
Six layers, one record underneath.
One confirmation updates every view at once, and the record improves the library behind the next answer. What is live is marked live — the rest is stated as what it is.
RecordLearner portfolio · every attempt, kept in orderLive
InsightCompletion journeys · repeated-mistake patternsLive
WorkspaceEditor · compiler · assignmentsLive · no device link yet
VerificationTeacher workspace · the confirmation gateLive
IntelligenceClass prep · photo diagnosis · guided questioningLive · one input of three
FoundationThe subject librariesOne kit, measured
The platform is identical in every subject. What changes is the library underneath it, so adding a discipline means adding a library — not rebuilding the system.
Programming
LiveEditor, compiler and checks. A compile result is deterministic.Electronics & robotics
LiveOne kit. One fault, shown detectable before it was offered.Mechanical & fabrication
Not startedNeeds a measured fault library before it can be listed.Lab sciences
Not startedNeeds a measured fault library before it can be listed.How it is confirmed
What decides is the input, not the confidence.
A device’s own report is a fact about the world. A model’s confidence is not — measured twice, it sat at 0.92–0.95 whether the diagnosis was right or invented. So the route is chosen by what produced the answer.
Step 1 · the learner submits their work
Provisional — they see it at once · not yet part of the record
Route A · v2
The compiler and its testsdeterministic — but nothing re-compiles yet, so this still goes to a personRoute B · v2
The device, over USB or Bluetooththe only input that can confirm itself — not shippedRoute C · live
The teacher who was in the roomevery result today takes this routeConfirmed · enters the record, and every view updates
What we have measured
A fault is not listed until it has been proven detectable.
Most of the accuracy in this product came from removing a claim, not adding a model. An input is only offered a fault once that input has been shown to tell the fault apart from its absence.
The rule
detectableVia is measured, not assumed. It reads like a description of the fault and is actually a claim about the model. The default for a new fault is an empty list, and it stays empty until a measurement fills it.
missing_resistorphoto · offered
led_reversed[ ] · photo removed
baud_mismatchusb_serial · unbuilt
led_reversed listed photo when the library was first written. Measured, it fired more often on correct boards than reversed ones — so the claim was removed rather than defended.
Why there are no figures here
The sweep was run against simulator screenshots. Printed beside the words “photo diagnosis”, a score would read as a claim about a classroom — which is the one condition it was not measured under.
Accuracy on real classroom photographs is unverified and recorded as an open question. It is the measurement that decides what this product is, and it has not been taken yet. An honest silence costs less than a number that means something narrower than it looks.
Our progress
What shipped, and what has not.
Dated, and marked with what it actually is. Anything not yet measured is listed as planned rather than delivered.
August 2026The workspace and compiler. Write, run and fix inside the platform, with every attempt kept as the completion journey.Shipped
August 2026Narrowed the electronics library after measurement. One detection claim did not hold up under test, so we removed it rather than defend it.Shipped
August 2026Photo diagnosis, grounded on the kit schema. One fault, shown detectable before it was offered, routed to a teacher for confirmation.Shipped
September 2026Real-hardware validation. Photo diagnosis has been tested against simulator screenshots, not classroom photographs. Until that sweep runs, accuracy on real photos is unverified.In progress
PlannedCode paste, enabled. The path is built and dark. It turns on the day a sketch fault has been shown detectable — measurement first, switch second.Not shipped
PlannedDevice diagnosis over USB and Bluetooth. The input that can confirm itself, and the only one that shortens a teacher's queue.Not shipped
Why we built it
The real meaning of study.
Study is how a person becomes able to do something they could not do before. That is the whole point of it, and it is what Kovaan OS is built to protect.
The trouble has never been that people stopped learning. It is that the learning happens somewhere nobody can see — in the four attempts before the one that worked — and every system we built throws those four away and keeps the fifth.
So keep the four. Let the learner watch themselves improve while it is still happening. Let the teacher see the mistake that keeps coming back, and fix the cause instead of marking the symptom. Let the institution see what its programme produced. Let anyone deciding about that person look at the work rather than a proxy for it.
That is what the next generation of education looks like to us. Not more content. Not more screens. A system where getting better is something everybody can see.
Ask
The things people ask first.
Short answers. Where something is not built yet, the answer says so.
Do learners need to install anything?
No. The editor, the compiler and the guidance run in the browser. Photo diagnosis works from a phone at the bench, which is where the learner already is.
Is keeping every attempt fair on learners?
The record is framed as effort, not as a penalty. Four attempts and a hint read as eleven minutes of work — and the number the learner is shown improving is how often they resolved something without asking.
What if the diagnosis is wrong?
It goes to the teacher who was in the room, which is why every result today takes that route. The model reports a confidence figure and it decides nothing — measured twice, that number sat in the same band whether the answer was right or invented.
How much can the photo input actually see?
Whether a component is in the circuit at all. Not which way round a part is fitted — that was tried, measured, and removed from the library rather than defended.
Is this a way of monitoring teachers?
No, and the reporting is framed to make that hard. Institution views are about the programme — where a cohort repeats itself. If half a class shares a weak point, that reads as a curriculum signal rather than a mark against the person teaching it.
Do we have to change our curriculum?
No. You keep your own module sequence and it stays private to you. Kovaan stores structure and technical parameters, never your lesson content.
Which subjects does this work for?
One kit today — Arduino Uno and breadboard basics — with a fault library that has been measured rather than assumed. A subject is added when its faults have been measured, not when its name has been typed in.
Who owns the data?
The institution and the learner. Personal identifiers are separated from diagnostic records, learners are minors so nothing real is held before formal consent, and an administrator can trigger full deletion.
Get started
Bring Kovaan to your programme.
Kovaan OS is pre-launch. We are running it with a small number of institutions first, so what you would be joining is a field test with the numbers published afterwards.
- 01A start dateAgreed with you, not assigned to you.
- 02Your programme set upCohorts, assignments and goals, entered once by us.
- 03A session run with youBefore a single learner sees it.
Talk to us
Sign-up is where the details are taken, so nothing typed here would be carried across. It is two screens and free to begin.
Get startedUsually answered the same dayFree to begin