How the AI marks an answer
When a teacher starts AI marking, the submitted files are prepared page by page, the writing is read, and each question is matched to the part of the script that answers it. Every question is then marked on its own against the rubric, and anything the AI cannot judge is set aside for the teacher.
How it is done today
Marking a hundred scripts by hand costs a teacher a week, and the last ones never get the attention the first ones did. A machine that puts a number on every question regardless is worse, because the teacher cannot tell where it was guessing.
How it works
- 1
Read the script
The submitted files are prepared page by page and their writing is read and stored, so every later step works from that text rather than the raw image.
- 2
Match questions to answers
Which part of the script answers which question is worked out first, so one question is never marked against the writing meant for another.
- 3
Mark each question on its own
Questions are marked one at a time, with the model given that answer's own region and the assignment's rubric criteria rather than the whole paper.
- 4
Subject knowledge where there is any
Where the subject has stored material, it is supplied alongside the question so the marking is done in the subject's own context.
- 5
Hand back what it cannot judge
A question the AI cannot settle is set aside for the teacher instead of guessed at, and the whole run is flagged as low confidence when that happens.
What the AI agent does, what you approve
What the AI agent does
Reading the script, matching answers to questions and proposing a mark per question against the rubric is the AI's work end to end.
What you approve
A proposed mark stays a proposal. What is recorded, and when the student sees it, are both the teacher's decisions.
What it looks like
Who uses it
- Used by:
- Teachers
- Institution types:
- Coaching centres
- Schools
- Colleges
- Private tutors
- Online teaching platforms
Plan availability
- Starter৳499 a month, one classroom
- Standard৳20 per student per month
- CustomNegotiated
Frequently asked questions
Does every question get a mark?
No. A question the AI cannot settle is set aside for marking by hand, and a single such question is enough for the whole run to be flagged as low confidence.
What if no question can be marked?
The run fails and says so rather than publishing a partial result. A half-marked script is never presented to the teacher as though it were complete.
Is the cost and time visible?
Each run records the tokens used, how many model calls it made and how long each stage took, so the cost of AI marking can be worked out rather than guessed at.
Can AI marking be switched off?
Yes. It can be paused from the platform side, and while it is paused a new run is refused with a message asking the teacher to try again later.