- Edge node
- NodeMCU multi-sensor publish
- Decision model
- SVM on moisture / temp / humidity
What it does
Timer-based irrigation ignores humidity and temperature. This project streams soil moisture, temperature, and humidity from a NodeMCU into Firebase Realtime Database, shows the feed in a Flutter app, and uses a small SVM to decide when to actuate a pump: with firmware hard cutoffs if cloud commands stop arriving.
Constraints
- Tiny edge RAM/CPU
- Unreliable Wi-Fi in the field
- Actuation still needs a local fail-safe
- Operators want live graphs, not only relay state
Decisions & tradeoffs
- Firebase as fan-out bus: many consumers without taxing the MCU
- SVM on three sensor features: cheap inference, easy retrain vs a neural net
- Client caching for dashboard continuity during brief drops
- Firmware emergency cutoffs independent of the cloud path
Tradeoffs: a cloud hop adds latency vs peer sockets and allows multi-consumer analytics. SVM fits low-dimensional sensor vectors; it will not magically generalize to unrelated climates without retraining.
What it demonstrates
- End-to-end IoT path from firmware to mobile UI
- Keeping safety on the edge when cloud control is optional
- Matching model complexity to feature dimensionality