AI script checker for handwritten exam scripts

Opening a session brings back its subject, its scripts and the state of each, with progress worked out from those states rather than guessed.

All the script-marking sessions you have opened, each with its state and how many scripts it holds. Draft, marking or finished, at a glance.

With no written rubric to hand, a structured one can be generated from the answer material, and the teacher edits it before any marking begins.

Each question, its marks and its choice rules are read out of the paper, including the condition familiar from our papers, answer any three of seven.

Scripts can be added to the same session at any time. Credits are charged per page rather than per script, and the whole batch is checked first.

Attach the marking scheme you already have and its per-question marks and mark bands are read straight out of it, with nothing to type in by hand.

Teachers place ticks, crosses and comments on the script page itself. Change a mark and the symbols on the page follow, so the two never diverge.

Where a paper offers a choice, the teacher can replace the automatic selection and name which answers count towards the total. Nothing is re-marked.

Script photos go straight from the phone to storage rather than through the server. Uploading costs no credits; those start when marking does.

Every question's marks are added up and checked against the paper's total. If they do not reconcile, the total is not published, however clean the rest.

When the handwriting changes partway through one script, or two students' pages end up together, it is flagged and the total is held back for review.

Which answers count towards the total is resolved from the paper's choice rules. Extra answers are still marked and shown, but not added in.

A page count that does not add up, writing cut off mid-word, a gap in numbering: five signs are weighed together, and one weak sign decides nothing.

Most flags are advisory, but any one of four serious ones stops the total being published at all until a teacher has looked at it and released it.

A student writing nothing and a page never being scanned are kept as different states here, rather than collapsed into one and the same zero.

Sideways phone photos are turned upright and large ones scaled down, but the scaling stops well above where Bangla conjuncts start to close up.

Each mark comes with how reliable it is. That figure is derived from outside checks rather than taken from what the model says about itself.

Every mark is checked against the bands the rubric defines, and a mark that does not fit is flagged in time to reach the verdict, not noted after it.

See at a glance how many more pages can still be marked. One credit is one page, so the balance reads directly as the number of pages remaining.

Open a session with the exam details, the question paper and the marking rubric. Creating it costs no credits; those are spent when scripts are added.

Choose a pack of 100, 500 or 1000 pages and pay through the SSLCommerz gateway. All three work out at the same rate, ৳0.30 per page, with no monthly bill.

Every proposed mark and comment can be changed on its own, and you can optionally record why: too harsh, too lenient, or marked against the wrong thing.

A script whose marking failed can be sent back to the queue. One still being marked cannot, because that would put two runs on the same script at once.

Upload scanned answer scripts and ClassTablet's AI proposes marks per question against your rubric. Teachers review and can override every mark.

Open any AI-graded script on one screen: page images, marks per question, the AI's reasons and warnings side by side, so a teacher can check each mark.