High-throughput multi-agent workflows optimized for ROCm and AMD Instinct GPUs, focusing on low-latency reasoning and scalable task automation.