Stop Guessing What's Important — Let AI Predict Your Exam Questions

The Exam Question Predictor is the app's past-paper analysis feature: upload the actual question papers your board or university has set in previous years, and the app reads them, works out which chapters actually get examined and how heavily, and generates likely questions for each one — chapter by chapter, in your own syllabus.

It lives on each subject's page and is part of Diya AI Pro. You can reach it two ways:

• The Upload Paper option from the syllabus menu.
• The Predicted Questions button that sits on a subject's card once you've opened it — before an analysis exists, that same button doubles as the invitation to run one.

From there:

• Upload up to 5 PDF papers, 5 MB each. Whatever past papers you can get your hands on — different years, different terms, mock papers from a coaching centre.
• Choose which subject(s) the papers cover. Normally one subject. But for a competitive exam — JEE, NEET, UPSC, admission tests where a single paper set spans Physics, Chemistry, and Maths in one sitting — you select all of them at once, and the AI sorts every question to the correct subject and the correct chapter on its own.
• Accidental overwrite protection: If a subject already has an analysis, running a new one warns you first and names exactly which subjects will be overwritten, so you don't lose a set of predictions by accident.

This is the part that makes the result more than a guess. Every analysis runs in two stages, and you keep both:

Stage 1 — Gemini reads the actual PDFs: The AI is handed your papers directly (not text extracted beforehand — the real PDF pages) along with your subject's actual chapter list, and asked to:

• Work out which chapter every single question in every paper belongs to.
• Score each chapter's importance from 0 to 100%, based on how often questions from it actually turn up across the papers you uploaded.
• Write 2–4 likely exam questions for every chapter that scored above zero.

This result appears almost immediately — you don't wait for the second stage to see something.

Stage 2 — A second AI cross-checks it, in the background: A different model (DeepSeek) is given the same extracted topics and independently does its own scoring and its own set of 2–4 questions, deliberately aimed at “deep reasoning and likely exam variations” rather than repeating the first pass. Once it finishes:

• Its questions are added alongside the first set — you don't lose either.
• The importance score you actually see is the average of the two AIs' scores, not just the first one. A chapter both models agree is heavily examined ends up clearly at the top; a chapter they disagree on lands somewhere in the middle instead of being decided by whichever model happened to run first.
• The chapter list re-sorts by this combined score the moment it's ready.
• In the app these show up as the Assistant Set (Gemini's questions) and the Professor Set (DeepSeek's) — you can flip between them per chapter, or view everything combined as All Questions. A few questions the AI genuinely can't tie to one specific chapter are kept too, rather than silently dropped, and still count toward the total.

Once generated, your predictions appear right where you study:

• On the chapter itself, right next to its title: Five importance bars that fill up like a phone's signal meter — from Not rated to Critical importance. These are scaled against the highest-scoring chapter in that subject's own analysis, so the bars always spread across the full range for that subject rather than clustering at one end if every chapter happens to score similarly in absolute terms.
• Under a chapter, when you expand it: The actual predicted questions for that chapter, with the Assistant Set / Professor Set toggle when both AIs produced results for it.
• On the subject card, one tap away: A summary button reading “Predicted Questions · N”, showing the total count and how many papers it was built from. Tapping it opens a full browsing sheet with the same three tabs — Assistant Set, Professor Set, and All Questions — so you can read through everything at once instead of chapter by chapter.

A syllabus in printed order treats every chapter as equally worth your time. Real exams don't. This is the tool that tells you which chapters your actual papers keep returning to — the ones worth an extra pass — versus the ones that show up once in five years.

It also quietly makes the rest of the app smarter: when the Study Planner generates a schedule with AI, it's told each chapter's past-paper weight as part of what it's working from, so a chapter your papers show up in nearly every year can be given more priority and more scheduled time than a chapter that's never once appeared — without you having to set that weighting by hand.

• Subscription required: Requires an active Diya AI Pro subscription — tapping in without one opens the subscription screen instead of the upload flow.
• File limits: Up to 5 papers per run, 5 MB each.
• Re-uploading: Running a new analysis replaces the old one for that subject — it isn't additive, so if you're adding a newly found paper to ones you've already analysed, re-upload the whole set together rather than just the new file.
• Chapter completion quizzes: One related but separate feature, so it isn't confused with this one: marking a chapter Completed can offer you a short multiple-choice quiz as a self-check. That quiz is generated fresh by the AI at the moment you finish the chapter — it isn't built from your uploaded past papers, and it exists whether or not you've ever run a paper analysis for that subject.