This tutorial has presented a high-performance AI Engine implementation of the popular MUSIC algorithm for estimating DOA using an antenna array. The MUSIC algorithm employed here adopts a QRD/SVD approach for subspace estimation that is well matched to the AI Engine compute capacity. Software pipelining techniques are employed to create a massively parallel data flow graph across \(157\) AI Engine compute tiles to implement a high-performance implementation capable of processing $128\times 8$ data snaphots at a sustained throughput rate of 1 $\mu s$ per snapshot. The full MUSIC algorithm including noise subspace basis identification, MUSIC spectrum evaluation, and null detection is implemented fully in the AI Engine array with no supporting logic from PL required. A comprehensive Hardware-in-the-Loop demonstrator system is built using an external host running Matlab to communicate to the VCK190 evaluation board over Ethernet.