33. Handle irregular data without losing the GPU
Design kernels for graphs, sparse matrices, irregular meshes, particles, and branching-heavy workloads. You will practice choosing layouts and algorithms that reduce divergence and wasted memory traffic.