A financial technology company is designing a system for real-time fraud detection. The system must process a continuous, high-volume stream of transaction data and identify suspicious patterns as they occur. A proposed architecture involves a large language model that, for each new transaction, queries a separate, large-scale database which is constantly updated with the latest transaction information to provide context. Critically evaluate this proposed design. What is the primary performance challenge this system would face, and why does this challenge make the architecture a poor fit for a streaming data application?