Showcase build lighter quantized model (int8). Per SA Admin this demo runs the edge lite weights on free tier compute. Full weights need dedicated GPU/TPU for higher fidelity.