This tutorial has presented the design of an MNIST ConvNet classifier in AIE-ML. The solution has ~100,000 parameters and requires ~18 tiles. It achieves a throughput of ~70K frames per second. The design illustrates many aspects of building “White-Box” designs for Machine Learning networks using the “dataflow” mode of Vitis AI Engine tool flows developed for signal processing. The similarity of code structure for the various layers of this simple design suggest a library of Machine Learning layers should be feasible in practice.