How would you design an intelligent, large-scale rate limiting system that uses machine learning to dynamically adjust request quotas based on user behavior and system health, ensuring high availability while mitigating sophisticated bot attacks?