Car accidents due to driver drowsiness is a very serious pan-world problem. At present there are few AI-based drowsiness detection and alert systems, but they are not suitable for cars due to their requirement of large memory, power, and latency and dependency on the cloud. Mammalian brain inspired Spiking neural networks, coupled with neuromorphic computing paradigm, can bring in a very low-power low-latency solution for the same. In this work, we have designed one such system and tested it on Brainchip Akida Neuromorphic hardware. We found that the system is highly accurate (92.01%) in identifying the drowsiness features of a driver. Moreover, latency and energy consumption of the system are found to be 46.5 ms/frame and as low as 16.2 mJ/frame respectively on hardware - which is pretty promising for deployment, especially in battery driven cars. In addition, we also showcase the on-chip learning capabilities of the system that let us learn new classes and subjects with minimal data and personalise our drowsiness detection system for a specific driver.