Аннотация
Home plant cultivation requires continuous monitoring and timely watering, which becomes especially challenging during owner absences. Current IoT irrigation systems suffer from three persistent limitations: dependency on continuous cloud connectivity, rigid pre-configured presets that are made for specific plants, and high technical barriers preventing adoption by non-expert users. This paper proposes a sustainable hybrid IoT architecture that resolves all three limitations simultaneously. The system integrates an ESP32 microcontroller with Large Language Model (LLM) configuration generation, an offline-first firmware architecture using non-volatile memory (NVM) storage, a Go and PostgreSQL backend REST API, and dual user interfaces via React web panel and Telegram bot. The LLM generates species-specific JSON watering configurations from a plain-text plant name, eliminating manual parameter entry. The device continues autonomous watering through network outages using locally stored configurations. Preliminary hardware testing on onion seedlings demonstrated successful automated watering execution. The proposed architecture represents the first documented application of generative AI for proactive IoT device configuration generation in the plant care domain, as confirmed by a systematic review of 21 recent publications (2023-2026).