kg-bootstrap

costiash's avatarfrom costiash

Creates and bootstraps Knowledge Graph projects from video transcripts. Extracts entities (people, organizations, concepts) and relationships into searchable graphs. Use when users want to build a knowledge graph, extract entities, analyze connections between people/organizations, or select Option 5 after transcription.

⭐4stars🔀0forks📁View on GitHub🕐Updated Dec 30, 2025
[Knowledge Base]

When & Why to Use This Skill

The Knowledge Graph Bootstrap skill automates the transformation of unstructured video transcripts into structured, searchable knowledge graphs. It intelligently identifies key entities—such as people, organizations, and concepts—and maps their complex relationships. By bootstrapping a custom schema and performing precise data extraction with evidence linking, it enables users to visualize and query the underlying connections within large volumes of spoken content.

Use Cases

  • Market Intelligence: Analyzing industry interview transcripts to map the network of founders, investors, and competitors within a specific sector.
  • Academic & Research Synthesis: Converting a series of educational lectures or seminars into a structured knowledge base of interconnected theories and citations.
  • Investigative Journalism: Tracking relationships and influence networks between public figures and organizations mentioned across multiple media sources.
  • Corporate Knowledge Management: Transforming internal meeting recordings into a searchable graph of project stakeholders, decisions, and cross-departmental dependencies.
namekg-bootstrap
descriptionCreates and bootstraps Knowledge Graph projects from video transcripts. Extracts entities (people, organizations, concepts) and relationships into searchable graphs. Use when users want to build a knowledge graph, extract entities, analyze connections between people/organizations, or select Option 5 after transcription.

Knowledge Graph Bootstrap

Step-by-Step Workflow

Step 1: Check Existing Projects

Use list_kg_projects first:

  • If projects exist → Ask: "Would you like to add to an existing project or create a new one?"
  • If no projects → Proceed to Step 2

Step 2: Create New Project (if needed)

Ask for a project name describing the research domain:

"What would you like to call this Knowledge Graph project?" Examples: 'Tech Industry Interviews', 'Climate Research', 'Company History'

Use create_kg_project and report the project ID.

Step 3: Bootstrap Domain Profile

This is where the magic happens:

Explain: "I'll analyze your transcript to discover entity types (people, organizations, concepts) and relationships."

Use bootstrap_kg_project with:

  • project_id: The newly created ID
  • transcript: Full transcript text (use get_transcript if needed)
  • title: Video/source title

Step 4: Present Bootstrap Results

Format results conversationally:

## Knowledge Graph Schema Discovered!

**Project:** [Name]
**Confidence:** [X]%

### Entity Types Discovered ([N])
- **Person**: Individuals mentioned
- **Company**: Organizations, startups
- [etc.]

### Relationship Types ([N])
- **works_at**: Employment relationships
- **founded**: Company founding
- [etc.]

### Example Entities Identified ([N])
- [Entity Name] ([Type])
- [etc.]

Now extracting entities into the graph...

Step 4.5: CRITICAL — Extract from Bootstrap Transcript

IMMEDIATELY after bootstrap, call extract_to_kg with the SAME transcript:

extract_to_kg(
    project_id="...",
    transcript="...",        ← Same transcript used for bootstrap
    title="...",             ← Same title
    transcript_id="..."      ← REQUIRED for evidence linking
)

This populates the actual graph with entities. Bootstrap only creates the schema (types and patterns); extraction adds the nodes and edges.

After extraction completes, present:

## Knowledge Graph Complete!

**Entities Added:** [N]
**Relationships Found:** [N]

Would you like me to:
1. Extract entities from another transcript
2. View current graph statistics
3. Explore key players and connections

Step 5: Continue Building

For subsequent transcripts on a bootstrapped project:

  • Use extract_to_kg directly (no re-bootstrap needed)
  • IMPORTANT: Always pass transcript_id to enable evidence linking
  • Show what was added vs. what already existed
  • Offer to continue with more transcripts

Tool Parameters

extract_to_kg (CRITICAL)

When calling extract_to_kg, always include:

  • project_id: Target KG project ID
  • transcript: Full transcript text
  • title: Video/source title
  • transcript_id: Required for evidence — Use the ID returned by save_transcript
extract_to_kg(
    project_id="abc123",
    transcript="...",
    title="Interview with CEO",
    transcript_id="5da6b612"  ← from save_transcript result
)

Without transcript_id, the graph inspector won't be able to show supporting evidence for entities.

Critical Rules

  • A project MUST be bootstrapped before extraction
  • Bootstrap happens ONCE per project (first transcript only)
  • Bootstrap creates schema ONLY — you MUST call extract_to_kg afterward to populate the graph
  • ALWAYS extract from the bootstrap transcript — don't skip to "another transcript"
  • Always pass transcript_id when using extract_to_kg
  • Always show progress in chat — user shouldn't need to check sidebar
  • The sidebar syncs automatically, but chat is the primary interface

Common Mistake to Avoid

❌ WRONG:

bootstrap_kg_project(transcript)
→ "Your KG is ready! Extract from ANOTHER transcript?"

✅ CORRECT:

bootstrap_kg_project(transcript)
→ "Schema discovered! Now extracting entities..."
extract_to_kg(same_transcript)
→ "X entities and Y relationships added!"