Six works.
Fourteen weeks.
One dimension that stopped.
KAUSHAL measures how a student in an art, craft or vocational programme is actually improving — one photographed work at a time. The instructor confirms every number. Nothing is recorded until they do.
The demo is a seeded academy with sixty-two students. It resets to a known state in one click, so nothing you press can break it.
Skill-based education has no gradebook.
In art, music, dance, craft and vocational training, students produce work rather than answers. There are no marks, no multiple choice, no submission log. Whatever the instructor knows about a student lives in the instructor’s head.
That works for eight students. It does not work for sixty. A plateau in one skill is invisible for months, and by the time it is obvious, the student has already decided they are not getting anywhere.
Students quit skill programmes not because they stopped improving, but because nobody could see that they had.
The whole product is three moves and a loop.
It has to fit in the two minutes after a class ends, or an instructor teaching sixty students will never use it twice.
- Beat 1
Photograph the work
At the end of the class, one photo per student. Under ten seconds each. It compresses on the phone before it uploads.
under 500 KB
before it leaves the phone - Beat 2
Confirm five numbers
KAUSHAL proposes a score and a reason for each of the five dimensions. You confirm, or you change one and confirm. Nothing is recorded until you do.
76567four inked, one still a guideline - Beat 3
Watch the line move
Every confirmed score lands on that student’s trajectory. When one dimension stops moving while the rest climb, KAUSHAL says so and writes the exercise that targets it.
Every academy has a different standard. The rubric learns yours.
A generic rubric is worse than no rubric, because it is confidently wrong in a way an instructor has to keep correcting. So the first thing KAUSHAL asks you to do is score ten of your own past works.
A work you scored at onboarding
Ten of these. You give each one the number your academy would give it.
line control
A foundation class and a portfolio class will not put this in the same place. Both are right about their own students.
The next work it scores
A different drawing, a week later. Your anchors go into the prompt with it.
what a generic rubric proposes
proposed against your standard
Because you place this work at 4, the proposal moved from 6 to 4. The rubric learned your standard, not a generic one.
Five dimensions you can point at on the paper.
There is no score for creativity, talent or effort in this product. A number nobody can verify by looking at the work would discredit the four next to it.
Proportion & placement
Are the relationships between parts, and their positions, accurate?
Line control
Is the mark-making confident, consistent and intentional?
Value range
How far apart are the darkest dark and the lightest light?
Edge quality
Is there variation between hard, soft and lost edges?
Composition
How well is the picture space used?
Built inside the problem, not next to it.
KAUSHAL is built by the person who runs Krishna Art Academy in Chennai — sixty-plus enrolled students, Udyam-registered. Every design decision in it answers a problem that shows up on a Saturday morning with thirty drawings on a table.
That is also why the scope is narrow. Calibration, scoring with confirm and override, the trajectory, plateau detection and the parent report are finished and work end to end. Nothing else was started.
- Where the numbers in this demo come from
- A seeded academy of sixty-two students, generated so the trajectory, plateau detection and cohort triage can be exercised honestly. The student records are synthetic; the workflow they run through is the real one.
- How plateau detection works
- We fit a straight line through each dimension's last 4 works. A dimension is flagged when its line is flat — under 0.06 points a week — while the other four are still climbing. No model, nothing hidden.
- What happens when the model is unavailable
- The image is measured in the browser — value spread, ink coverage, edge variance, mass distribution — and those measurements produce a labelled fallback score with low confidence. The product degrades to something honest rather than to a spinner.
- Who owns a score
- The instructor. Every proposal is stored separately from the confirmed number, overrides are kept, and no unconfirmed score is ever presented as fact anywhere in the product.
KAUSHAL cannot grade art, and does not claim to.
A model looking at a photograph of a drawing can tell you how far apart the darkest and lightest values are, and it can be wrong about almost everything else. On an ambiguous work — a deliberate stylistic choice, an unusual medium, a bad photograph — its proposals are unreliable, and it says so by lowering its own confidence.
So the product never puts a number on a student without a human agreeing to it. What KAUSHAL genuinely does is the bookkeeping a person teaching sixty students has no time to do: it remembers every score, draws the line, and tells you which part stopped moving.