DeepTutor — AI-Powered Personalized Learning Assistant

Introduction

DeepTutor is an open-source AI learning assistant developed at the University of Hong Kong that turns any document, textbook, or research paper into an interactive tutoring session. It combines multi-agent reasoning with retrieval-augmented generation to provide adaptive explanations, practice questions, and deep research capabilities tailored to each learner.

What DeepTutor Does

• Processes uploaded PDFs, papers, or textbooks into a knowledge base for interactive Q&A
• Uses multi-agent orchestration to break down complex topics into digestible explanations
• Generates practice questions and quizzes based on the source material
• Performs deep research across uploaded documents to answer nuanced questions
• Adapts explanation depth based on the learner's demonstrated understanding

Architecture Overview

DeepTutor employs a multi-agent architecture where specialized agents handle different aspects of the tutoring process. A coordinator agent routes user questions to retrieval, reasoning, or generation agents as needed. The retrieval layer uses vector embeddings and keyword search to find relevant passages, while the reasoning agent chains multiple LLM calls to build step-by-step explanations. The system persists conversation history and learning progress for continuity across sessions.

Self-Hosting & Configuration

• Requires Python 3.9+ with pip for dependency management
• Supports OpenAI, Anthropic, and local LLM backends via configurable provider settings
• Documents are indexed locally using FAISS or Chroma for vector storage
• Web interface runs on Gradio with configurable port and authentication
• Environment variables control API keys, model selection, and chunk sizes

Key Features

• Upload any PDF or academic paper for instant interactive tutoring
• Multi-agent deep research that synthesizes answers from multiple document sections
• Socratic teaching mode that guides learners through problems step by step
• Concept mapping that visualizes relationships between topics in uploaded materials
• Session persistence for long-term learning progress tracking

Comparison with Similar Tools

• NotebookLM — Google's document AI; DeepTutor adds multi-agent reasoning and deeper pedagogical features
• Quivr — RAG framework for documents; DeepTutor specializes in tutoring with adaptive explanations
• Kotaemon — Document chat interface; DeepTutor provides structured learning paths beyond simple Q&A
• ChatPDF — Simple PDF chat; DeepTutor offers multi-document research and quiz generation

FAQ

Q: What file formats does DeepTutor support? A: It primarily supports PDF documents, with community extensions for EPUB and plain text files.

Q: Can I use local LLMs instead of API-based models? A: Yes, DeepTutor supports Ollama and other OpenAI-compatible local endpoints.

Q: Is DeepTutor suitable for university-level coursework? A: Yes, it handles research papers and technical textbooks with multi-step reasoning for complex topics.

Q: How does it handle math and equations? A: DeepTutor renders LaTeX equations from PDFs and can explain mathematical derivations step by step.

Sources

• https://github.com/HKUDS/DeepTutor
• https://arxiv.org/abs/2501.09891