NLU in virtual assistants converts user utterances into machine interpretable representations, key challenges include ambiguity, out of domain queries, and real-time latency