Millions of people around the world are affected by Epilepsy, and Sudden Unexpected Death is commonly attributed to generalized tonic-clonic seizures. Effective management requires timely detection and intervention, which remain challenging with current wearable solutions relying on single-modal sensors and cloud-based processing. These systems suffer from latency, privacy concerns, and limited predictive capabilities. We propose an IoT-based wearable system integrating multimodal sensors: accelerometer ACC, gyroscope GY, magnetometer MG, electrocardiogram ECG, electrodermal activity EDA, photoplethysmography PPG and infrared temperature TMP; and a lightweight TinyML model deployed on an Nvidia Jetson Nano. This architecture enables real-time inference directly on the edge device, ensuring low latency, privacy preservation, and energy-efficient operations. Moreover, a selective data collection mode saves resources and facilitates the generation of high-quality biosignal datasets for model refining. The proposed system addresses critical gaps in existing solutions, offering a scalable, privacy-preserving, and responsive framework for real-time seizure detection and future predictive capabilities. The findings provide opportunities for improved healthcare for patients and more effective epilepsy management by utilizing multimodal sensing on the edge.