What is dual-model architecture in classification systems?

Dual-model architecture uses two separate models for complementary classification tasks such as candidate generation and ranking.

In large-scale systems, a single model is often insufficient for both speed and accuracy. A dual-model architecture separates responsibilities: a fast lightweight model generates candidates (recall-focused), and a slower, more complex model ranks or refines predictions (precision-focused). This separation improves scalability and reduces computational cost while maintaining high accuracy.