Hey, I've been thinking about the trade-offs between computational efficiency and explainability in AI models. How do you balance that in your designs?