How do you analyze system logs to identify latency issues in a live conversational AI pipeline at Interface Ai?