Given a large, complex codebase, what is your strategy for debugging an inference latency issue in an edge ML system like Syntiant?
