About ExamShuffle
ExamShuffle turns one exam into several, so the paper on each desk asks the same questions in a different order.
Why I built it
I am a software engineer, and my wife is a teacher. I saw how much of her time went into writing several versions of the same exam, then making sure every version had an answer key that actually matched it. None of that work improves the exam. It only multiplies it, and one slip in a key means marking a whole stack of papers wrong.
ExamShuffle is my attempt to take that burden off educators. Instead of spending hours producing versions and checking keys by hand, you upload the exam you have already written and get the versions back, each with its own correct key, in a few minutes.
Why cheating matters to me
I also built it because I dislike cheating. An exam is only useful if it shows what each student actually knows, and copying quietly takes that away: from the teacher trying to see who needs help, and from the students who did the work honestly. Different papers on neighbouring desks will not end cheating on their own, but they make copying harder and less rewarding. If a tool can make that the easy default rather than an evening of extra work, it can help move the culture in the right direction.
How the shuffling works
Questions and answer choices are reordered per version, with each answer key written from its own paper. What moves, what is held in place and why is covered on How it works.
What it does not do
It does not write questions, mark submissions, or host your exam. It does one step — the rearranging — and hands the documents back to you. Scanned documents with no text layer cannot be read, because there is nothing in them to reorder.
Your documents
The original file is never saved, and the questions read from it are deleted after eight hours. The privacy policy has the detail.
What comes next
The weakest step today is reading the document. Questions are found by rules that look for numbering and lettered options, which works well on a neatly formatted exam and stumbles on unusual layouts. What I am planning, though it is not built yet, is to add a language model to that reading step, so that more formats are understood correctly the first time and there is less to fix on the review screen. If it ships, the privacy policy will be updated first to say exactly what is sent and where.