Highlighting a PDF Isn't Learning: DeepTutor Uses Quizzes to Make You Actually Learn

Highlighting a PDF is not learning; a week later, you'll forget most of it. DeepTutor takes the opposite path. Hand it your materials—books, notes, YouTube lectures—and it turns around and teaches you.

The project comes from the HKUDS team at the University of Hong Kong. It is open source, has 42.9K GitHub stars, runs locally, and uses your own model.

What it does:

• After reading the content, it quizzes you immediately
• It doesn't just give you the answer; it walks you through the problem step by step
• Ask questions while watching YouTube lectures, and it answers from the corresponding timestamp
• Generates study plans, tracks what you have actually mastered, and exposes weak points

Some online commenters joked that 'master's degrees are finished,' while others put it more directly: 'Quizzing is more effective than highlighting—every time.'

The logic behind it is active recall. Highlighting is passive underlining; quizzes force the brain to retrieve information, and the latter leaves a much deeper memory. DeepTutor turns this into a product: open source, locally deployed, with data never leaving your device.

The installation method is also worth noting. The easiest way: paste the GitHub link into Claude Code or Codex and have it run 'install and set up DeepTutor for me.' The repo includes an installation guide written specifically for AI agents; the agent can configure the environment and launch the web app by itself.

This is also more than just a chatbot. Chat, Quiz, Research, Visualize, and Mastery Path share the same runtime and session context; the knowledge base supports RAG engines such as LlamaIndex, GraphRAG, LightRAG, and PageIndex; there is also an EduHub marketplace where community skills can be imported.

For learners, it turns materials into a teacher. For developers, it shows how an agent-native architecture can be made into a concrete educational product. The repo is on GitHub.