MasterClass

AI-driven tutoring, interview prep, and research system powered by Claude Code.

Three specialized agents — Tutor, Interviewer, and Researcher — guide you through course planning, interactive lessons, mock interviews, and paper reading, all via slash commands in your IDE.

Quick Start

npm install
npm run install:mc      # interactive setup — picks languages, personas, paths

The installer will:

1. Collect your preferences (name, languages, agent personas)
2. Compile agent profiles and workflow files into .masterclass/
3. Generate IDE slash commands in .claude/skills/
4. Set up a Python venv with embeddings and vector search dependencies

Once installed, open Claude Code and run /mc-start to begin.

Slash Commands

Command	Description
/mc-start	Initialize session and select an agent
/mc-course-plan	Create or adjust your course roadmap
/mc-lesson-plan <path>	Plan a specific lesson
/mc-lesson-teach <path>	Interactive teaching session
/mc-interview-run <path>	Run a mock coding interview
/mc-paper-read <path>	Read a paper section by section
/mc-paper-survey	Survey papers on a topic
/mc-paper-qa <path>	Q&A about a specific paper
/mc-kb-search <query>	Search the knowledge base
/mc-kb-build	Build or rebuild knowledge base indexes
/mc-context	View or update your learner profile
/mc-settings	View and modify configuration
/mc-help	Detect project state and suggest next steps

Agents

Agent	Default Name	Role
Tutor	Alex	Course planning, lessons, practice exercises
Interviewer	Jordan	Mock interviews with structured code review
Researcher	Morgan	Paper reading, literature surveys, research Q&A

Each agent has a configurable persona style (e.g. patient / strict / socratic for the tutor). Run /mc-settings to change.

Project Layout

src/
  engine.xml              # Core execution rules
  agents/                 # Agent persona templates
  workflows/              # Workflow definitions (markdown)
  actions/
    ts/                   # TypeScript actions (config, compilation)
    py/                   # Python actions (embedding, vector search)
scripts/
  install.ts              # Interactive installer
docs/                     # Generated — course content and progress
data/
  knowledge/              # Papers, books, leetcode problems, notes
  .index/                 # FAISS vector indexes (auto-built)
.masterclass/             # Generated — compiled runtime artifacts
.claude/skills/           # Generated — IDE slash commands

docs/, data/, and .masterclass/ are gitignored — they hold your personal content and compiled artifacts.

Knowledge Base

Drop files into data/knowledge/ by category:

• papers/ — PDFs (with optional -notes.md companions)
• books/ — PDFs
• leetcode/ — Markdown problem sets
• notes/ — Free-form markdown

Run /mc-kb-build to index, then /mc-kb-search <query> to search. Uses FAISS + BM25 hybrid search with local HuggingFace embeddings (bge-m3 or bge-small-en, chosen at install time).

How It Works

• File-as-state: All progress lives in file content — no database, no external state. Empty file = not started; has content = resumable.
• Sequential workflows: Each workflow is a series of markdown steps executed one at a time by the LLM engine.
• LLM-as-judge: The interviewer evaluates code with a structured rubric (correctness, complexity, quality, edge cases, communication).
• Multilingual: Communication language, document output language, and interview language are independently configurable.

Requirements

• Node.js >= 18
• Python >= 3.10 (for knowledge base features)
• Claude Code CLI or IDE extension