voidful/hung-yi-lee-skill

What it solves

It transforms a vast collection of educational content—including hundreds of YouTube videos and transcripts—into a structured "skill" for AI assistants. Instead of a simple persona prompt or basic RAG, it captures the specific teaching methodology, knowledge framework, and linguistic style of Professor Hung-yi Lee, allowing an AI to explain complex AI concepts using his unique pedagogical approach.

How it works

The project uses a multi-step "distillation" process to build a comprehensive knowledge base:

1. Linguistic Analysis: It analyzes transcripts to identify specific speech patterns and frequent phrases.
2. Knowledge Graph Construction: It extracts 916 concept nodes and over 3,600 edges from course materials to map how concepts relate to one another.
3. Human Calibration: The developers interviewed Professor Lee to correct misconceptions and refine the AI's rules (e.g., ensuring the AI prioritizes context and punchlines over dry lists).
4. Structured Output: The AI follows a specific "DNA" for explanations: starting with intuition, moving to a "black box" overview, explaining the internal mechanism, warning about pitfalls, and ending with a recap.

Who it’s for

• Learners who want to study AI concepts using Professor Hung-yi Lee's intuitive teaching style.
• AI developers looking for a sophisticated way to implement expert personas beyond simple prompting.
• Users of AI coding assistants who want to integrate a specialized knowledge base into their workflow.

Highlights

• Knowledge Graph: Features a detailed graph with 916 nodes and 10 concept communities (e.g., ML Fundamentals, Agents, Diffusion).
• Evidence-Based Style: Speech patterns are derived from frequency analysis of actual transcripts rather than impressions.
• Traceability: Answers can be traced back to specific lectures and timestamps.
• Pedagogical Framework: Implements a strict teaching sequence (Intuition $\rightarrow$ Black Box $\rightarrow$ Mechanism $\rightarrow$ Pitfalls).