C4 Model - Level 1: System Context
System Context Diagram
System Context Narrative
Primary Actor: Content Creator
The Content Creator is the main user who initiates the pipeline by providing a video URL. This could be:
- A software engineer wanting to document technical talks
- A researcher processing lecture videos
- A technical writer generating documentation from demos
Secondary Actors
Knowledge Worker
Consumes the generated artifacts (executive summaries, glossaries) for:
- Quick understanding of video content
- Research compilation
- Knowledge base building
Developer
Uses technical artifacts (SDD, TDD, ADR, C4 diagrams) to:
- Understand system architecture
- Implement similar patterns
- Generate project documentation
External System Relationships
| External System | Purpose | Data Flow |
|---|---|---|
| YouTube/Vimeo | Video source | Out: URL, In: Video file + metadata |
| OpenAI API | Speech-to-text transcription | Out: Audio, In: Transcript |
| Anthropic API | Vision analysis, LLM generation | Out: Images/Prompts, In: Analysis/Text |
| Serper API | Web search for context expansion | Out: Queries, In: Search results |
System Boundary
The Video-to-Knowledge Pipeline encapsulates:
- Video download and preprocessing
- Audio extraction and transcription
- Frame sampling and deduplication
- Vision-based content analysis
- Content synthesis and chunking
- Multi-format artifact generation
- Document inventory management
The system does NOT include:
- Long-term video/audio storage (ephemeral processing)
- User authentication/authorization
- Real-time collaboration features
- Custom ML model training
Key Scenarios
Scenario 1: Technical Talk Processing
- User provides YouTube URL of conference talk
- System downloads video (15 min, 1080p)
- Extracts audio → transcribes to 2,500 words
- Samples 450 frames → deduplicates to 120 unique
- Analyzes unique frames for diagrams/code
- Generates SDD, TDD, ADR, C4 diagrams
- Outputs structured documentation package
Scenario 2: Tutorial Video Processing
- User provides Vimeo URL of software tutorial
- System processes video with screen recordings
- Higher deduplication threshold for static screens
- Emphasizes UI component detection
- Generates step-by-step outline + architecture docs
Data Sensitivity
| Data Type | Sensitivity | Handling |
|---|---|---|
| Video content | Medium | Ephemeral, deleted after processing |
| Transcripts | Low | Stored as output artifacts |
| API keys | High | Environment variables only |
| Search queries | Low | May be logged for debugging |