How would you optimize a model using quantization, pruning, or TensorRT/ONNX compilation for edge deployment at Anduril Industries?