DeepTutor Skill for Claude Code

A Claude Code skill that brings DeepTutor's graph-enhanced RAG tutoring methodology to your terminal. Transform documents (PDFs, textbooks, papers) into interactive tutoring sessions with knowledge graphs, dual-loop problem solving, deep research, and adaptive question generation.

Credits & Attribution

This skill is a Hybrid Claude-Native adaptation of DeepTutor by the HKU Data Science Lab (HKUDS).

The original DeepTutor is a full-stack AI tutoring platform (FastAPI + Next.js) with 6 agent systems. This skill adapts its methodology and pedagogical patterns -- not its code -- into a lightweight Claude Code skill that leverages Claude's native reasoning capabilities.

Key patterns adapted from DeepTutor:

• Dual-loop problem solving (Analysis Loop + Solve Loop)
• Graph-enhanced RAG retrieval (entity-relationship knowledge graphs)
• Investigate Agent's precision query methodology
• Research pipeline with topic decomposition
• Difficulty-calibrated question generation
• Citation-driven answering

If you use this in academic or professional work, please cite the original DeepTutor project:

@software{deeptutor,
  author = {HKUDS},
  title = {DeepTutor: AI-Powered Personalized Learning Assistant},
  url = {https://github.com/HKUDS/DeepTutor},
  year = {2024}
}

What It Does

Workflow	Description
Initialize KB	Process a document into chunks + knowledge graph
Dual-Loop Solve	Analysis (gather context) then Solve (step-by-step reasoning)
Generate Questions	Practice questions calibrated by difficulty
Deep Research	Multi-section academic reports with citations
Guided Learning	Prerequisite-aware step-by-step teaching
KB Management	List, inspect, explore, delete knowledge bases

Installation

Quick Install

# Clone this repo into your Claude Code skills directory
git clone https://github.com/ndpvt-web/deeptutor-claude-skill.git ~/.claude/skills/deeptutor

# Install the only dependency
pip install networkx

Manual Install

# Create the skill directory
mkdir -p ~/.claude/skills/deeptutor/scripts

# Copy files
cp SKILL.md ~/.claude/skills/deeptutor/
cp scripts/*.py ~/.claude/skills/deeptutor/scripts/

# Install dependency
pip install networkx

Usage

Once installed, the skill is automatically available in Claude Code. Trigger it by:

• Saying deeptutor or /deeptutor
• Asking to "study", "learn from", or "understand" a document
• Asking to "create a knowledge base" from a PDF
• Requesting "practice questions" or "quiz me"
• Asking to "explain" or "solve" a textbook problem
• Requesting a "deep research report"
• Asking for "guided learning" on a topic

Example Workflows

Study a PDF:

"deeptutor - help me study this textbook" (with a PDF attached)

Solve a problem:

"Using my ML knowledge base, explain how backpropagation works step by step"

Generate questions:

"Quiz me on Chapter 3 of my textbook - medium difficulty"

Deep research:

"Write a deep research report on regularization techniques from my notes"

Architecture

~/.claude/skills/deeptutor/
  SKILL.md                    # Skill definition (triggers, workflows, prompts)
  scripts/
    kb_manager.py             # Knowledge base CRUD operations
    graph_builder.py          # Knowledge graph (NetworkX) construction
    graph_retriever.py        # Hybrid retrieval (BM25 + graph expansion)
  kb/                         # Knowledge bases stored here (auto-created)
    <kb_name>/
      metadata.json           # KB metadata
      chunks/                 # Document chunks
      graphs/                 # Knowledge graph JSON

How It Works

1. Claude handles: PDF reading, text understanding, entity extraction, reasoning, tutoring, question generation, report writing
2. Python scripts handle: Graph storage (NetworkX), BM25 scoring, graph traversal, knowledge base persistence
3. The SKILL.md orchestrates: Workflow definitions, prompt patterns, tool usage instructions

This division keeps the skill lightweight (~27KB total, 1 pip dependency) while capturing 70% of DeepTutor's value.

Requirements

• Claude Code (with skill support)
• Python 3.9+
• networkx package (pip install networkx)

Comparison with Original DeepTutor

Aspect	Original DeepTutor	This Skill
Dependencies	50+ packages	1 package (networkx)
Infrastructure	FastAPI + Next.js + DB	Claude Code + 3 Python scripts
LLM Provider	OpenAI/Anthropic API	Claude (native)
Graph Engine	RAG-Anything/LightRAG	NetworkX + BM25
Footprint	500MB+	~27KB
Features	100%	~70%
Setup time	10+ minutes	< 1 minute

License

MIT License. See LICENSE.

The original DeepTutor project has its own license -- please check their repository for details.

Acknowledgments

• HKUDS/DeepTutor -- The original AI tutoring platform whose methodology this skill adapts
• HKU Data Science Lab -- For their research on graph-enhanced RAG and adaptive tutoring
• NetworkX -- Graph library used for knowledge graph operations

---

Keywords

For the search engines and the curious humans who find things by typing random words

AI tutoring, knowledge graph tutoring, RAG tutor, adaptive learning AI, AI study assistant, intelligent tutoring system, graph-enhanced learning, research paper tutor, AI education tool, Claude Code tutor skill, deep learning tutor, study helper AI, academic AI assistant