Kb Ready Extraction

Knowledge Base-Ready Extraction (KB-Ready Extraction) Skill

This skill implements the KB-Ready Knowledge Engineer persona and architectural framework. It is specifically designed to transform raw speech-to-text transcripts (ASR), lecture recordings, messy meeting notes, or unstructured web content into structured, high-density, Knowledge Base (KB) and RAG-ready Markdown documents.

Scope: Knowledge distillation, conversational denoising, information structuring, and RAG chunk optimization. Target Systems: Open-Notebook, Obsidian, Logseq, NotebookLM, Notion, Local RAG Vector DBs (Chroma, Qdrant, SurrealDB). Complementary to SMART POLE: While structures prompts & tasks (Task Specification), structures learning & factual content (Knowledge Representation).

How to Load This Skill

1. Set system prompt: Load references/system-prompt.md as the agent's system prompt (paste into Custom GPT Instructions, Gemini Gem, Claude Project, or automated extraction pipeline).
2. Provide reference files (optional but recommended): Make references/schema-specification.md, references/denoising-rules.md, and references/rag-optimization-guide.md available as context.
3. Execute extraction: Feed raw transcript text along with metadata (Course name, Module, Lesson title). The agent outputs a single, clean, valid Markdown document with YAML Frontmatter.

Reference Files

The 4-Stage Knowledge Engineering Pipeline

┌────────────────────────────────────────────────────────┐
│ 1. INGESTION & CONVERSATIONAL DENOISING               │
│    - Strip filler words (à, ừm, thì, là, mà)           │
│    - Strip mic checks, greetings, ASR repetition loops │
└────────────────────────┬───────────────────────────────┘
                         ▼
┌────────────────────────────────────────────────────────┐
│ 2. SEMANTIC DISTILLATION                               │
│    - Extract definitions, frameworks, formulas, steps  │
│    - Capture Case Studies & Anti-Patterns (Mistakes)   │
└────────────────────────┬───────────────────────────────┘
                         ▼
┌────────────────────────────────────────────────────────┐
│ 3. PEDAGOGICAL STRUCTURING                             │
│    - Inject YAML Frontmatter (Metadata & Summary)      │
│    - Organize into H1, H2 (Key Takeaways), H3 sections │
└────────────────────────┬───────────────────────────────┘
                         ▼
┌────────────────────────────────────────────────────────┐
│ 4. RAG & SEARCH OPTIMIZATION                           │
│    - Synthesize 3-5 Reverse QA Pairs (HyDE alignment)  │
│    - Validate YAML delimiters & Heading boundaries     │
└────────────────────────────────────────────────────────┘

The Golden Document Structure

Every document generated by this skill strictly follows this canonical layout:

---
title: "{Clear, descriptive lesson/article title}"
course: "{Course / Category / Book title}"
module: "{Module or Topic cluster, or empty}"
instructor: "{Instructor / Author name, or Unknown}"
difficulty: "{Cơ bản | Trung cấp | Nâng cao}"
target_audience: "{Who this knowledge is for}"
tags:
  - "{tag 1}"
  - "{tag 2}"
summary: "{Concise 2-3 sentence summary of the core knowledge}"
---

# {Lesson / Note Title}

## 💡 Tóm Tắt Cốt Lõi (Key Takeaways)
- {Key takeaway 1}
- {Key takeaway 2}
- {Key takeaway 3}

## 📖 Nội Dung Chi Tiết
### 1. {Main Section 1}
{In-depth academic/practical explanation with bullet points and bold technical terms.}

> **Ví dụ / Case Study:** {Real-world examples, numbers, or practical applications}

> ⚠️ **Lỗi Sai Thường Gặp (Anti-Pattern):** {Common mistakes, misconceptions, or what to avoid}

### 2. {Main Section 2}
...

## ❓ Câu Hỏi & Trả Lời Trọng Tâm (Q&A for Search)
- **Hỏi:** {High-intent natural question matching user search intent?}
  - **Đáp:** {Direct, comprehensive, self-contained answer.}
- **Hỏi:** {Question 2}
  - **Đáp:** {Answer 2}
- **Hỏi:** {Question 3}
  - **Đáp:** {Answer 3}

Why This Structure Maximizes RAG & Knowledge Bases

1. High Signal-to-Noise Ratio: Eliminates 100% conversational debris, maximizing embedding semantic density.
2. Deterministic Chunk Boundaries: Splitters split cleanly on ## and ### without severing contextual sentences.
3. HyDE Reverse Retrieval: The ## ❓ Q&A section matches user query vectors with high cosine similarity.
4. Actionable Anti-Patterns: Capturing mistakes/pitfalls prevents downstream AI agents from suggesting bad advice.
5. Metadata Filtering: Frontmatter fields enable hybrid search (metadata filtering + dense vector search) in Open-Notebook, Chroma, and Qdrant.