A system with two separate neural networks—one that processes questions and one that processes documents—both converting their inputs into comparable vector embeddings.