Skip to main content

Vision, Core Tenets & Bilingual Strategy

The DOCCAD Vision​

DOCCAD envisions an engineering organization where documentation is not an afterthought, but an active, verifiable asset that continuously models reality.

Canonical GitHub Documentation -> Project Knowledge -> AI Processing ->
{ User Docs | Technical Architecture | Recruiter Views } -> Interview Preparation

User Question -> Retrieve Project Knowledge -> Analyze Context ->
Generate Dedicated Page -> PR Review & Persistence

What Exists Today vs Vision​

To preserve transparency, DOCCAD maintains a strict distinction between what is designed, what is proven, and what is planned:

  1. Proven & Verified: A static publishing framework (Docusaurus 3.x), two-plane structural separation, deterministic Level-1 retrieval, hash-based drift detection, task contracts, and complete offline generation via deterministic fixtures.
  2. Designed for Production: The thin provider abstraction with adapters for Anthropic, Google Gemini, OpenAI, and local Ollama; GitHub Actions CI/CD workflows for PR-gated persistence.
  3. Deferred to Scale: Vector database retrieval (Level 2/3), automatic translation of dynamic derived views, and enterprise SSO.

The Six Non-Negotiable Tenets​

  1. Files over Databases: GitHub is the primary system of record. Every architectural decision, schema, prompt, and documentation page is a versioned file.
  2. AI in CI, Never in Serving: Ordinary readers never interact with runtime AI models. Published pages are statically built HTML/JS assets with zero runtime LLM costs or single points of failure.
  3. Canonical Independence: Canonical documentation can be built, published, and read even if every AI provider on earth is unreachable.
  4. Governed Derivation: AI-generated views enter the repository solely through human-reviewed pull requests with provenance metadata.
  5. Deterministic Mechanical Drift: Staleness is tracked through sha256 content digests, not fragile timestamps or imprecise AI guesswork.
  6. Bilingual EN/HU Architecture: English is the authoring and canonical language. Key landing, overview, and onboarding materials are localized into Hungarian (hu), with transparent indicators when deep technical specifications remain English.