QuizForge

Everyone studying for an exam has the same problem: a pile of material and no good way to test themselves on it. Re-reading feels productive but barely works; self-testing is what actually builds recall.

QuizForge closes that gap. Paste in your notes, a textbook section, or upload a text file, and QuizForge produces a complete study session:

A study guide — the material's key concepts, each tied to the sentence that explains it.
A real quiz — multiple-choice questions with plausible distractors, plus short-answer questions, generated from your material only.
Instant grading with explanations — miss a question and you immediately see the correct answer and the source statement it came from.
A spaced-repetition review plan — the exact topics you missed, scheduled for review at +1, +3, and +7 days, when memory science says they'll stick.

The demo uses an excerpt from a real first-responder exam study guide: load the sample, forge the quiz, and watch it build questions about AVPU, SAMPLE, OPQRST, and normal vital-sign ranges.

How AI is used: QuizForge asks a large language model (OpenAI's open-weight GPT-OSS 120B, served on Groq's free tier; an OpenRouter free model is the second choice) to read the study material and return a strict JSON study package: guide entries and validated questions with topics and explanations. Every AI-produced question is machine-validated before it reaches the learner — option counts, answer-index ranges, and required fields are checked, and malformed output is discarded rather than shown.

And because a study tool that breaks is worse than no tool, QuizForge ships with a built-in offline engine: if no API key is configured or the provider is unreachable, a local heuristic engine (key-term extraction + cloze question construction) generates the quiz from the text itself. The interface honestly labels which engine forged each quiz.

Built fresh for this hackathon: QuizForge was designed and written from an empty folder during the submission period, per the contest rules — scope, PRD, and spec documents are in the repo alongside the code, and every line of code carries a plain-English comment explaining what it does.