Add TechGraph on Google
Speaking with TechGraph, Karan Gupta, Co-Founder of AssessPrep, discussed how schools are becoming increasingly cautious about handing grading decisions to AI even as the administrative burden of assessments continues to grow, and how the company is addressing that challenge through a human-in-the-loop model where AI drafts a score and rationale while teachers retain final authority over every evaluation.
Gupta also highlighted how this approach has helped schools reclaim more than one million teacher hours by automating repetitive assessment tasks rather than replacing educators, where analytics drawn from more than five million student submissions now help teachers catch a struggling class or a quietly slipping student long before a final exam would have shown it.
Read the interview in detail:
TechGraph: For years, assessments have largely measured what students remember at a particular moment in time. With AI becoming part of the classroom, do schools need to rethink what assessments are designed to measure rather than simply digitising traditional examination models?
Karan Gupta: Yes, and it’s a conversation we have with coordinators almost every week. Putting a paper exam on a screen isn’t transformation. It’s the same test with a new delivery method. The real question is what the test is actually measuring.
Recall still matters, but if a student can look up a fact in seconds, memorising it is no longer the skill worth grading. What’s worth measuring is whether they can apply an idea, analyse it, and defend it. IB and Cambridge frameworks have valued that for years, but timed paper exams struggle to capture it.
So we built AssessPrep as an on-screen assessment platform, not a way to digitise the same old paper. When a teacher can drop in a source document, an audio/video file, a simulation, or a multi-part problem, the assessment starts testing how a student thinks, not just what they’ve memorised. That’s what schools should be designing for.
TechGraph: AI can reduce the administrative burden of creating and grading assessments, but consistency and fairness remain equally important. How does AssessPrep ensure that automation strengthens teachers’ judgement rather than replacing the professional discretion that educators bring to evaluation?
Karan Gupta: Our rule is simple: AI does the heavy lifting and the teacher takes the decision. It’s a human-in-the-loop design, and it sits behind every feature we ship.
Take grading, for example: our AI can draft a score and a rationale against the rubric in seconds. But the teacher sees that suggestion and has the ability to edit it, override it, or reject it. Nothing reaches a student without a human signing off. Across our schools, that’s helped reclaim over a million teacher hours, not by removing teachers, but by removing the drudgery around them.
Consistency actually improves this way. Human marking naturally drifts when you’re tired. The last script of the night rarely gets the same fresh attention as the first. AI removes that drift by giving every script the same rubric-based starting point, and the teacher then applies the judgement only they have. The mechanical part gets automated and, most importantly, the discretion stays with the educator.
TechGraph: Students now have easy access to AI tools that can generate answers in seconds. How should schools rethink assessment design so that academic integrity depends less on surveillance and more on evaluating genuine understanding?
Karan Gupta: Surveillance alone is a losing game. You can lock the browser, and we do offer a secure browser and lockdown mode for the high-stakes exams where that genuinely matters. But you can’t monitor your way to real understanding. Integrity has to be designed into the assessment, not just policed around it.
The most effective thing schools can do is set tasks AI can’t easily fake. Ask students to reason from a specific data set, respond to a source they’ve just been given, show their working, or defend a position on the spot. When the question demands the student’s own thinking, a generated answer stands out straight away.
The goal isn’t catching cheaters; it’s designing assessments where genuine understanding is the only way through.
TechGraph: AssessPrep works with schools managing different curricula, teaching styles, and assessment frameworks. Across this diverse customer base, what common challenges do educators face when integrating AI into assessment workflows, and where does resistance tend to emerge?
Karan Gupta: The challenges rhyme across curricula, whether it’s IB DP, MYP, Cambridge, or Edexcel. First is trust. A teacher who’s marked their subject for fifteen years is right to be skeptical of a machine suggesting a grade. Second is workload fear: is this one more system I have to learn? Third is fairness. Nobody wants to explain to a parent that a score came from a black box.
Resistance usually shows up at the grading stage, not question-creation. Teachers happily let AI help build a question bank or draft a quiz. They get cautious the moment it touches a student’s mark, and they should.
What moves them is control and transparency. Once teachers see they can review and override every AI suggestion, and see why the AI scored something the way it did, scepticism turns into adoption. The resistance isn’t to AI. It’s about losing control.
TechGraph: Assessment platforms now generate far more data than traditional examinations ever could. Beyond reporting marks, how do you see assessment analytics shaping teaching strategies, student interventions, and curriculum planning over the next few years?
Karan Gupta: This is the part that excites us most, because a mark on its own tells you almost nothing. The analytics tell you the story behind the mark.
At the classroom level, a teacher can see that three-quarters of the class missed the same sub-skill, and reteach it tomorrow instead of finding out at the final exam. At the student level, patterns surface early, so the quiet student quietly falling behind gets flagged before it becomes a crisis. At the coordinator level, you see which topics underperform across cohorts and where the syllabus needs more time. With over five million student submissions on the platform, those patterns are rich.
Over the next few years, assessment stops being the thing at the end of learning. It becomes a live feedback loop that shapes teaching while there’s still time to act.
TechGraph: Schools are increasingly expected to adopt AI responsibly while addressing concerns around transparency, bias, and student data. What principles have shaped AssessPrep’s approach to ensuring teachers remain in control of AI-assisted decisions rather than becoming dependent on them?
Karan Gupta: A few principles have guided us from the start. First, the teacher is always the final decision-maker. AI recommends, humans decide. We deliberately don’t grade a student without a teacher’s sign-off.
Second, transparency over magic. If our AI suggests a score, it shows the reasoning against the rubric. A teacher should never have to trust a number they can’t interrogate.
Third, schools stay in control of the AI itself. They choose which features to switch on. It’s not all-or-nothing. A school can adopt AI authoring but keep grading fully manual, or roll features out slowly as trust builds. And we handle children’s data across 85+ countries with the seriousness that demands. Schools’ data belongs to schools.
The biggest safeguard isn’t using less AI; it’s keeping the teacher’s judgement switched on, so an educator stays sharper and faster, never a passive approver clicking accept.
TechGraph: Looking ahead, do you expect examinations to remain the primary way of measuring student performance, or will continuous AI-supported assessment gradually become the new standard across schools?
Karan Gupta: It’s not either/or, and we’d be cautious of anyone who says exams are dead. High-stakes examinations aren’t disappearing. Systems still need a common, secure benchmark.
What changes is everything around that final exam. The big terminal test stops being the only signal and becomes the last data point in a much richer, continuous picture. Instead of a student’s year hanging on three hours in a hall, teachers have months of evidence: where the student grew, where they stalled, what they can actually do. This is exactly what India’s NEP points to as well, a shift away from one-shot, high-pressure exams toward formative, continuous assessment.
AI is what makes that practical, because constant assessment used to buckle under the marking load, which is the exact burden automation lifts. So exams stay, but they stop standing alone. The new standard is ongoing, teacher-led, AI-supported assessment, with the formal exam as one chapter, not the whole book.