Carlota Weather Station

Carlota is a weather station reporting live from the house it is named after. It measures temperature, humidity and atmospheric pressure, and publishes them on a dashboard that needs no login.

Estação Meteorológica Carlota — dados de ambiente em tempo real.

Why it exists

IA.PURU’s argument is that owning the data collection is what makes applied AI worth anything. Carlota is that argument reduced to its smallest working form: a sensor we chose, on hardware we flashed, writing to a database we run, feeding a page we built. Every layer is ours, and the whole path is short enough to inspect.

It is also the first production deployment of Beacon Manager — the platform stopped being a fleet-management abstraction and started having an actual device to manage.

How it works

The sensor — a BME280 wired over I2C at address 0x76 to an ESP32-S3 running the Beacon firmware.
The device — polls the platform for its configuration, reads the sensor on the interval it is told to use, and POSTs each reading to the backend.
The store — Fastify writes the readings to PostgreSQL through Prisma, timestamped both by the device and on arrival.
The dashboard — a Next.js app rendering three current-value cards plus time series over 24 hours, 48 hours and 5 days, charted with Mantine Charts on Recharts.

The dashboard never talks to the backend directly from the browser: a server-side route proxies /api/environment, so the internal backend URL stays internal. There is no authentication on the public app, deliberately — the weather is not a secret.

What I did

Wired and mounted the station, and flashed the firmware;
Added BME280 support and the public environment-data endpoint to the Beacon platform;
Designed the dashboard, its time windows and its server-side proxy;
Deployed it behind Nginx with PM2 on the IA.PURU infrastructure.

Check it out at https://carlota.iapuru.com.br