UVERD

The sensors sit in the rainforest. The network does not. Readings had to survive intermittent connectivity, arrive in volume, and still be queryable fast enough that a dashboard feels live, all while staying legible to conservation staff, who are field people and not data analysts.

MQTT over Mosquitto on the ingest path, so devices buffer and reconnect instead of quietly dropping readings. PostgreSQL with TimescaleDB behind it, because the entire access pattern is time-series and an ordinary table stops being fast somewhere around the millionth row. A Svelte front end that opens on a map and drills down to a single device, with thresholds that raise an alert rather than expecting someone to sit and watch a chart.

Illegal logging shows up as an alert instead of a satellite photo months later. Pollution indicators are tracked continuously. Millions of data points are stored and queried, and the people using it every day did not need training to read it.