Building a bounded AI guide for process-based learning
How I used an append-only event log, deterministic progression, bounded context, semantic validation, and educator-authored fallbacks to keep an LLM inside a deliberately narrow role in a learning application.
Read engineering post