By , splitting it into streaming segments , leveraging the Edge TPU for hardware‑accelerated decoding , and **optimising the Kisslick UI and Fantasia model