SeraFullAutonoma

Farming has to deal with labour shortages, with using water and energy carefully, and with producing food sustainably.

The goal was simple: a greenhouse that monitors itself, makes its own decisions and talks naturally with the people who use it.

An ESP32-based layer reads air and soil conditions, light, water level and rain, and controls irrigation, a retractable roof and a mobile rover with a camera.

A manager agent built with CrewAI and LLMs coordinates agents for greenhouse operations, planning and business management. They reason over live and historical data instead of fixed rules. Computer vision finds plant diseases and weeds in images taken inside the greenhouse.

A local wake word, speech-to-text and text-to-speech pipeline lets you ask the greenhouse to open the roof, start irrigation or report on the plants. The web and mobile app, in React, TypeScript and Capacitor, covers live monitoring, history, manual control, inventory and energy. The backend runs on FastAPI, PostgreSQL and Docker, behind an HTTPS reverse proxy.

At Unihack 2025, our team built the first version in 48 hours and won the Grand Prize.

I defended it as my bachelor’s thesis at UPT’s Faculty of Automation and Computers and received the highest grade, 10/10.

The prototype is deployed at serafullautonoma.ro.