Self-healing distribution systems are one of the foundational requirements for future smart grids that are built to withstand disturbances, accommodate bidirectional power flow, and also maintain reliability despite the threat of in- creasing renewable penetration. Traditional FLISR (Fault Location, Isolation, and Service Restoration) solutions used currently depend mostly on SCADA, PMUs, and other high- cost protection relays. This infrastructure is usually not un- available in low-voltage networks, microgrids, and academic environments for teaching purposes. Our work proposes a novel low-cost, microcontroller-based self-healing grid pro- totype that uses ACS712 current sensors, ESP32/ESP8266 wireless sensing nodes communicating via ESP-NOW, and an STM32 Nucleo 64 (F446RE) microcontroller executing an embedded Random Forest classifier through the Eloquent- TinyML library. This system automatically and autonomously detects, classifies, and isolates faults based on a real-time multi- feature current signature. Our experimental setup and further validation shows an overall classification accuracy of 92.76%, ESP-NOW latency of 12-to 18 ms over 22 metres, and a pro- tection response time under 200 ms. Compared to other con- ventional schemes, our proposed architecture provides an inex- pensive yet robust platform similar to SCADA-like self-healing behaviours.