SwitchBraidNet is a quantisation-aware EEG classification architecture that achieves high accuracy in motor imagery and SSVEP tasks. It outperforms four baselines in FP16 and FP32, with MI accuracy of 69.49%, SSVEP accuracy of 93.48%, and a hybrid information transfer rate of 64.82 bits/min in FP16. The model runs efficiently with only 3.03 KB of INT8 storage, enabling low-power embedded deployment.