Building an Agentic IoT System with Gemma 4 and ASHA for Plant Care

This article details a project submitted to the Gemma 4 Challenge, demonstrating how small open-weight AI models can manage Internet of Things (IoT) systems. The author developed ASHA, an ecosystem providing a natural language interface between AI models and physical hardware. The prototype, named Cinderella, monitors a house plant using low-cost components including an ESP32 WROOM microcontroller, water level sensor, light strip, and buzzer, totaling under $25. The system allows users to interact with the device via WhatsApp, enabling commands such as checking water levels, controlling lights, or setting autonomous care schedules without complex dashboards. A key finding was the model's reliability; when the microcontroller was deliberately shut down, Gemma 4 correctly identified the tool timeout as an error rather than hallucinating data. This demonstrates the potential for cost-effective, responsive AI-driven IoT solutions that avoid expensive cloud inference latency.