Engineering Decisions

Multiple specialized AI agents communicate through an IPC bus, with an orchestrator routing queries to the appropriate agent based on intent classification.

Lightweight edge devices (ESP32) handle sensing and actuation, while compute-heavy tasks (ML inference) run on an edge server. Communication is via MQTT.

A single repository containing web, mobile, desktop, and API projects sharing TypeScript types, utilities, and business logic through internal packages.

All services (app, database, cache, reverse proxy) run as Docker containers orchestrated by docker-compose with health checks, resource limits, and automated backups.

Multiple ML algorithms (XGBoost, LightGBM, Random Forest, Neural Network) make independent predictions, which are combined by a meta-learner for superior accuracy.

Mobile applications cache data locally (SharedPreferences, SQLite, or Firestore offline mode) and sync when connectivity is restored. Essential for IoT apps where internet may be unreliable.