This work presents a microcontroller-based system for detecting Denial-of-Service (DoS) attacks in IoT environments using edge AI. A lightweight neural network model, trained with Keras and optimized through quantization, runs in real-time on an Arduino Portenta H7 using TensorFlow Lite for Microcontrollers. Unlike previous approaches, traffic analysis is performed on a Linux host, which extracts packet features and sends them to the microcontroller via a serial-over-TCP connection. A custom Python pipeline manages communication. Experimental results show that, despite its limited resources, the Portenta H7 can classify network flows as benign or malicious in real-time, demonstrating the feasibility of hybrid edge AI solutions with low power consumption.