Pipeline
The Doc Generator operates as a four-stage pipeline with iterative refinement.Architecture Overview
Stage 1: Conceptual Design
LLM Role: Senior Technical Architect Temperature: 0.3 (consistent, structured output)Output
JSON document blueprint including:- Document architecture with section purposes
- Section depths and word count estimates
- Key concepts and explanation approaches
- Comparison frameworks with evaluation criteria
- Audience profile and learning objectives
- Success criteria for quality assessment
- Research requirements
Stage 2: Draft Generator
LLM Role: Senior Technical Writer Temperature: 0.5 (balanced creativity and consistency)Output
Full Markdown document with:- Clear hierarchical structure (H1, H2, H3)
- Technical depth appropriate to audience
- Code examples where applicable
- Diagrams described in Mermaid syntax
- Cross-references and links
Stage 3: Quality Review
LLM Role: Senior Editor / Tech Lead Temperature: 0.2 (precise evaluation)Scoring Dimensions
| Dimension | Description |
|---|---|
| Technical Accuracy | Facts, code, and concepts correct |
| Clarity & Structure | Logical flow, readable prose |
| Completeness | All required topics covered |
| Audience Fit | Appropriate depth and tone |
| Actionability | Reader can take action |
Decision Logic
Stage 4: Finalization
LLM Role: Document Publisher Temperature: 0.1 (deterministic formatting)Tasks
- Generate YAML frontmatter
- Apply consistent formatting
- Add metadata (timestamps, version)
- Prepare for GitHub commit
API Throttling
Built-in delays prevent rate limiting:| Between | Delay |
|---|---|
| Conceptual → Draft | 3 seconds |
| Draft → Review | 3 seconds |
| Review → Revision | 5 seconds |