Skip to content

Knowledge Base Status

Current focus

AI Foundations and the full Agentic Engineering sequence are complete at first-draft depth. The current documentation pass has also normalized the published knowledge base into English and Hungarian variants so the site no longer alternates languages between chapters.

Current state

  • The repository is private and knowledge/ remains renderer-independent.
  • Published knowledge pages are maintained as English default files (name.md) with Hungarian siblings (name.hu.md).
  • Material for MkDocs provides English/Hungarian language switching plus system/light/dark appearance switching.
  • GitHub Actions validates strict MkDocs builds and deploys successful main pushes to Rackhost over SSH.
  • AI Foundations is complete with 14 detailed first-draft topics in both languages.
  • Agent Skills is complete with 12 detailed first-draft topics in both languages.
  • Agentic Loops is complete with 12 detailed first-draft topics in both languages.
  • Agent Architecture is complete with 12 detailed first-draft topics in both languages.

Decisions

  • English is the stable/default knowledge-document filename; Hungarian uses the .hu.md suffix.
  • User-facing knowledge changes should keep both language variants aligned.
  • Treat agentic applications as normal software systems with explicit deterministic boundaries around probabilistic behavior.
  • Keep canonical state, authorization, budgets, approval, side-effect guarantees and lifecycle outside the model.
  • Build model context as a purpose-specific projection of authoritative state/evidence rather than a hidden database.
  • Organize modules around business capabilities and bounded contexts instead of AI providers/frameworks.
  • Use modular monolith, Hexagonal Architecture, Clean Architecture, dependency inversion and event-driven boundaries where useful.
  • Keep LLM providers, vector/search stores, MCP/connectors, queues and external integrations behind explicit ports/adapters where practical.
  • Prefer deterministic workflows with bounded agentic islands where overall control flow is known.
  • Use separate agents only for real responsibility, permission, trust, scaling or ownership boundaries.
  • Use retrieval for reference knowledge and live application/tool access for current operational state.
  • Treat security, reliability, evaluation and observability as core architecture responsibilities.

Next steps

  • Review/refine the completed Agentic Engineering chapters while studying them.
  • Keep EN/HU variants synchronized as the material evolves.
  • Add practical implementation exercises/architecture examples where useful.
  • Choose the next AI workstream: RAG, MCP or Memory.
  • Verify/enable HTTPS when the domain/DNS setup is ready.

Area status