We treat “is this answer real?” as a research question.
Most AI study tools hope you don't check. We score every generated answer against the source it came from, publish the methodology, and let the citation do the talking. A formal research collaboration with a Canadian university is in development.
Faithfulness scoring
Every generation is scored against the source with ROUGE-L, BERTScore, and an LLM-judge entailment check. Results land on the public leaderboard.
Retention measurement
FSRS review outcomes feed back to generation quality so we can answer: did variant X actually help students remember more in 30 days?
Hand-graded corpus
Academic source documents with expert-written summaries, flashcards, and span-level citation annotations, released under CC-BY-SA when the corpus stabilizes.
Citation-first generation
The model cannot emit a claim without a structured citation. Ungrounded claims are refused at the schema level, not hidden.
Built by the customer it's for.
I'm a Software Engineering student at Ontario Tech. I built WizzStudy because I needed it. Every feature you see started as a problem I had at my own desk: a 200-slide deck two days before an exam, a recorded lecture I missed half of, a flashcard set that worked for week 3 but not week 8.
That's why citations are non-negotiable, FSRS replaces SM-2, rooms have a tutor that knows the source, and the recap audio uses the actual lecturer's key phrases. None of it is decoration.
Faculty or graduate student?
We're actively recruiting a faculty academic supervisor and a graduate research intern at a Canadian university.