Ended up building a full relationship intelligence tool: - Ambiguous message decoder (22 "fine" variants mapped) - Communication pattern analysis - RAG-powered response drafting - Error types with severity and recovery suggestions
The architecture treats communication as a domain problem. Messages have context/tone/timing, patterns aggregate into signals, errors have root causes and recovery paths.
Tech: Effect-TS, SQLite, LanceDB. All local, privacy-first. No cloud.
Demo at lifeops.in
Curious what HN thinks about applying systems thinking to fuzzy human problems.