Skip to main content

DOCCAD — GitHub-Native AI Documentation System

Welcome to DOCCAD, an open-architecture, GitHub-native documentation ecosystem designed for high-assurance engineering teams. DOCCAD transforms version-controlled canonical project knowledge into governed, evidence-backed derived views.

The Core Philosophy​

Traditional technical documentation suffers from two failure modes:

  1. Manual Drift: Architecture documents, runbooks, and API guides slowly fall out of sync with actual code implementations.
  2. AI Hallucination & Pollution: Ungoverned AI tools inject unverified assertions, imaginary metrics, and deprecated practices directly into source repositories.

DOCCAD eliminates both problems by establishing Two Structurally Separated Content Planes:

  • The Canonical Plane (/docs): Human-authored, reviewed, and immutable to automated generation. This is the single source of truth.
  • The Derived Plane (/views): AI-generated views (such as recruiter summaries, interview guides, and dedicated question answers) produced strictly from canonical evidence through PR-gated workflows.

Key System Tenets​

  • Docs-as-Code: Everything is a version-controlled file in Git. No runtime database, no vector database, and no proprietary content silos.
  • AI-Free Static Serving: The published documentation is a static website (Docusaurus 3.x). Readers and search engines access documentation with zero runtime AI latency, cost, or outage vulnerability.
  • Deterministic Level-1 Retrieval: Context assembly uses exact file paths, frontmatter dependency closures, and keyword scanning. Retrieval is deterministic, reproducible, and verifiable in pull requests.
  • Hash-Based Mechanical Drift: Every generated view records the sha256 content hash of each canonical file used as evidence. When a source file changes, CI flags only affected derived pages as stale.
  • Anti-Overengineering: The platform is designed to be operated and maintained by a single capable engineer.

Architecture at a Glance​