Difficulty-Aware Routing

A dynamic task-assignment mechanism that automatically selects between a low-cost rule engine and a high-performance large language model based on the complexity of each input sample. This approach maintains accuracy while controlling inference costs, enabling scalable deployment.