Every exam a school gives starts the same way: a blank page, and a teacher who has to fill it with questions that are fair, that match what was actually taught, and that still feel fresh in the fifth year of teaching the same syllabus. Our exam paper builder is built for that specific evening — the one before a test is due, when a teacher would rather be marking the last stack of homework than writing question eleven from scratch.
The evening before the exam
Writing a good exam paper takes longer than most people outside a classroom realise. It is not enough to think of questions; they have to cover the right proportion of the syllabus, sit at the right difficulty for the grade, avoid repeating last term's wording too closely, and still be marked by a rubric that holds up under scrutiny from parents and inspectors alike. Multiply that by every subject, every grade, every term, and it becomes one of the least visible but most time-consuming parts of a teacher's year.
Faced with that workload, the honest shortcut many teachers take is to reuse an old paper, or lightly edit one from a previous batch of students. It is a sensible response to real time pressure, but it has a cost: questions leak between year groups, papers stop reflecting what was actually taught this term, and the exam drifts slightly further from the classroom each time it is recycled. We wanted to give teachers back the evening, without asking them to give up a fresh, accurate paper in exchange.
What the paper builder actually does
A teacher picks a grade, a subject, and the syllabus or curriculum the school is actually following, and the builder generates a full set of questions aligned to that combination — not generic questions pulled from nowhere, but ones scoped to what that cohort has covered. A teacher can steer the spread: more weight on a chapter the class struggled with, a mix of short-answer and long-form questions, a particular difficulty curve for a weaker section versus a stronger one.
- Questions are generated against the grade, subject, and syllabus the school actually teaches — not a generic bank
- A teacher can ask for more or less emphasis on specific chapters or topics
- The output is a draft paper, not a finished one, presented for review before anything is used
- Every question can be edited, swapped for another, or rewritten entirely by the teacher
That last point is the one we treat as non-negotiable. The builder produces a draft. It is the teacher who decides what actually goes in front of students — accepting a question as written, adjusting the wording, changing the marks it carries, or discarding it altogether in favour of something they write themselves. Nothing reaches a student's desk without a teacher having looked at it first and put their name behind it, in the same way they always have.
A first pass on grading, not a final one
The same principle carries into grading. Once answers come in, the platform can offer a first read of a student's response — a suggested mark, and a short note on what the answer got right or missed. This is deliberately framed as a starting point rather than a verdict. A teacher marking forty scripts still has to read each one, but instead of starting from a blank rubric every time, they start from a draft they can accept, adjust, or override outright.
The mark that actually reaches a student's record is always the one the teacher confirms, never one issued automatically the moment an answer is scanned. If the suggested read misjudges a partial answer, or misses that a student solved a problem in a valid but unconventional way, the teacher corrects it, and that correction is what stands. The AI's job here is to save the time spent on the first read-through, not to make the decision that the read-through exists to produce.
Why exams are held to a stricter standard
We apply this same assist-not-replace approach across the platform, but we hold the line more firmly here than almost anywhere else, because an exam paper and its grading are among the highest-stakes artifacts a school produces. A homework suggestion that misses the mark costs a student a slightly less useful comment. A wrongly calibrated exam question, or a grade that reaches a permanent record without a teacher actually having signed off on it, can affect how a student is placed, how they see their own ability, and what opportunities open or close for them later. That is not a place for a system to guess and move on.
AI can shorten the mechanical, repetitive parts of exam work — drafting questions, giving a first read of an answer. The judgement of whether a question is fair, and whether a mark is earned, stays with the teacher, by design, not as an afterthought.
— PDP Shikshya product notes
A live debate, and where we come down
This sits inside a debate that is very much alive right now, well beyond any one platform or country. Educators, policymakers, and parents around the world are actively working through what it means for AI to grade a student's work, or generate the questions that work is measured against. The concerns raised are legitimate: whether an automated first pass on an answer is genuinely as fair and attentive as a teacher's own read, whether bias creeps into how responses are scored, and whether AI-written questions can really be calibrated to a specific curriculum, a specific class, and where that class actually is in its learning — rather than to some generic average student who does not exist in any real classroom.
We do not think that debate gets resolved by promising AI will do the whole job better than a teacher, nor by refusing to use it at all and leaving teachers to absorb every hour of drafting and marking by hand. The middle position — and the one this feature is built around — is to let AI take on the parts of the work that are genuinely mechanical and repetitive, while keeping the parts that require judgement, context, and accountability firmly with the person who is trained for them and answerable for them. A teacher who has taught a class all term knows things about that cohort no model can see: who is coasting, who is anxious under exam conditions, who solved a problem the unconventional but correct way. That knowledge is exactly what the sign-off step is designed to protect.
What sign-off looks like day to day
In practice, the workflow is straightforward. A teacher requests a paper for a grade, subject, and syllabus scope, reviews the draft that comes back, edits or replaces whatever does not fit, and only then finalises it for use. On the grading side, each answer arrives with a suggested mark and a short rationale attached, which the teacher reviews alongside the actual script before confirming, adjusting, or overriding it. In both cases, the AI output is visibly a draft state within the platform — it does not quietly become final on its own, and a teacher always has to take an explicit action to make it so.
This also means the tool improves the more a teacher uses it thoughtfully. Editing a suggested question or adjusting a suggested mark is not extra overhead layered on top of teaching; it is the same professional judgement a teacher already exercises, just applied to a faster starting draft instead of a blank page or an unmarked script.
The line we keep coming back to
None of this is about making exams less human. It is about giving back the hours that used to go into the most repetitive parts of writing and marking them, so a teacher has more of the evening left for the parts only they can do — deciding whether a question is genuinely fair to this class, and whether an answer truly earned the mark it received. The paper builder and the grading assist are tools for that evening, not a replacement for the teacher sitting through it.