Upload past papers, the rubric and your notes. ExamTwin reconstructs the real exam and compiles the highest-value timed rehearsal for the time you have today.
Four questions. One scored diagnosis. Then the next rehearsal recompiles.
Take a real four-question slice. ExamTwin scores it, explains every answer and compiles what you should rehearse next.
Your materials become an exam blueprint. Every attempt changes what the engine builds next.
Replay three attempts. Each result changes the evidence, the readiness signal and the next rehearsal ExamTwin compiles.
Change the exam and watch the rehearsal recompile around its actual pressure.
The same source material can produce more noise or one clear next step.
Generation is only the last step. First comes a reconstructed exam, real evidence and a deliberate constraint.
ExamTwin started with a team of students, a hackathon deadline and a frustration: plenty of study content, but no system for deciding what to practise next.
Read our story →No. ExamTwin first reconstructs and verifies the exam blueprint. Questions are generated only inside a timed rehearsal chosen for a specific evidence goal.
Past papers, the official rubric or syllabus, and your own notes. The more faithful your source material, the closer the reconstructed exam is to the real thing.
Yes — that's the point. You tell ExamTwin the timebox you actually have today, and it compiles the single highest-value rehearsal that fits it.
Every attempt updates the evidence: which skills are weak, how close the exam is, and where you lose points. The next rehearsal is recompiled from that, not repeated.
A small team of students in Vienna. We started it at OpenAI Build Week 2026 because it's the study tool we needed ourselves.
Upload the material. Get one focused rehearsal with a reason behind it.
Build my exam twin →