AI & Exams

How AI Predicts Exam Questions with 85% Accuracy

By the StudyClaw Team · Feb 28, 2026 · 5 min read

If you've ever looked at a past year paper and thought "this same topic came up 4 times in 5 years," you've discovered the secret behind exam prediction. StudyClaw does this systematically, at scale, using Gemini AI.

The Problem with Studying Everything

Most engineering syllabi have 5–6 units per subject, each with 10–15 sub-topics. That's 60–90 topics to potentially study — but a 3-hour end-sem paper typically tests only 10–15 concepts. The students who score 80%+ don't study everything; they identify which topics always show up.

💡 GTU Pattern Insight: In Circuit Theory, "Thevenin's Theorem" has appeared in 9 out of the last 10 GTU end-semester papers. StudyClaw flags it with 90% probability.

How StudyClaw's PYQ Engine Works

Our predictor runs in three stages when you ask "predict questions for Circuit Theory":

1. Historical Scraping — The pyq_scraper.go module crawls our internal archive of GTU past papers from the last 5 years. It tags each question with a subject, unit, and core topic.

2. Frequency Analysis — The pyq_predictor.go module computes how often each topic appears across years, weighted towards recent papers (which carry more weight in trending topics).

3. AI Synthesis — The ranked frequency data is sent to gemini-2.5-flash with a focused prompt: "Given these topic frequencies, predict the 5 most likely exam topics, accounting for the new GTU syllabus."

📊 Example output for "Circuit Theory":

1. Thevenin's Theorem — 90% probability
2. Node Analysis — 75% probability
3. AC resonance circuits — 65% probability
4. Superposition Theorem — 60% probability
5. Mesh Analysis — 55% probability

Why "85% Accuracy"?

In our informal testing across 3 GTU semesters, 8–9 of the top 10 predicted topics appeared in the actual paper. This isn't magic — it's pattern recognition. University exam setters tend to rotate through the same core concepts.

The Caveats

This is a prediction, not a guarantee. New syllabus changes can shift patterns significantly, which is why the engine explicitly focuses on the "current syllabus format." Always study the syllabus fully — use predictions to prioritize, not to skip.

Try it yourself: open your StudyClaw WhatsApp bot and send: Predict exam questions for [your subject]