Quizzer

Quizzer turns PDFs into interactive, Duolingo-style courses you can study from the terminal.

It combines PDF ingestion, question generation, spaced repetition, adaptive difficulty, and lightweight gamification so a static document becomes something you can actually practice.

Built on the five pillars of the Duolingo Method: learn by doing, personalized learning, focus on what matters, stay motivated, feel the delight.

• Generate full courses from textbooks, manuals, lecture notes, or handouts
• Study with seven question types, from easy recognition to free recall
• Revisit the right material at the right time with SM-2 scheduling
• Chat with an in-context tutor after each question
• Keep momentum with XP, streaks, and lesson unlocking

Features

• PDF to course -- Drop in any PDF and generate a full course with 7 question types (true/false, multiple choice, matching, sorting, fill-in-the-blank, ordering, free recall)
• Spaced repetition -- SM-2 algorithm schedules reviews so you see material right before you'd forget it
• Adaptive difficulty -- Tracks your ability and picks questions in your zone of proximal development
• Scaffolded learning -- Questions progress from receptive (true/false) to productive (free recall) within each session
• Gamification -- XP, streaks, levels, and LLM-generated effort-based praise
• Tutor chat -- After any answer, type /chat to open a multi-turn conversation with the LLM about the question, with the source page images as context
• Review mode -- Interleaves due questions across completed lessons, with mistakes weighted 2x
• Any LLM backend -- Works with any OpenAI-compatible API endpoint

Requirements

• Python 3.14+
• An OpenAI-compatible LLM endpoint (local or remote)

Installation

Install from PyPI:

pip install quizzer-ai

Install from source:

git clone https://github.com/suncloudsmoon/quizzer.git
cd quizzer
pip install .

Quick start

quizzer

On first launch, Quizzer prompts you for an OpenAI-compatible endpoint, API key, and model name. After that, these commands drive the flow:

During a study session, after answering a question:

How it works

1. Course generation (/create): Renders each PDF page as an image, sends it to a vision-capable LLM to triage and summarize content, then generates questions with calibrated difficulty. The result is a read-only course.json.

2. Study sessions (/study): Builds a session from due questions, filtered by adaptive difficulty and scaffolded from easy/receptive to hard/productive. After each answer you rate your confidence; the system updates your spaced repetition schedule and ability estimate accordingly. You can then type /chat to discuss the question with an LLM tutor that has the source page images as context.

3. Mistake resurfacing: Incorrect answers resurface at the end of the session for a second attempt.

4. Lesson progression: Lessons unlock linearly. Achieving 70%+ accuracy completes a lesson and unlocks the next.

5. Review: Once multiple lessons are complete, review mode interleaves due questions across all of them.

Why it feels different

Most PDF study tools stop at summarization. Quizzer is built around active recall instead:

• It asks you to answer, not just read
• It adapts question difficulty to your current ability
• It brings mistakes back before they disappear from memory
• It keeps each lesson moving from recognition toward recall
• It lets you open a tutor chat exactly when confusion happens

Example setup

This project has been tested with LM Studio using the lmstudio-community/ministral-3-8b-reasoning-2512 model for question generation. Note that course generation can take a long time depending on your hardware -- for a long book (~1000 pages), expect 12 hours or more.

Data storage

Quizzer uses platformdirs to store files in OS-standard locations:

Typical platform directories:

Project structure

src/quizzer/
  app.py              CLI entry point, config, LLM setup
  quiz_creator.py     PDF -> course JSON (offline, run once per document)
  quiz_handler.py     Serves questions at runtime
  learner_state.py    Per-learner runtime state (single JSON file)
  paths.py            Platform-specific paths via platformdirs

Dependencies

• Pydantic -- data models
• LlamaIndex -- LLM interface and structured output
• PyMuPDF -- PDF page rendering
• platformdirs -- platform-specific config/data directories
• tqdm -- progress bars during course generation