An LLM agent must classify support tickets and route them to billing, technical, or account teams. Which architecture is MOST appropriate?

The task requires classifying an incoming request (the support ticket) to determine its nature and then routing it to the correct specialized handler (the billing, technical, or account team). The most appropriate architecture is one that first performs intent classification to understand the user's goal. Following classification, a routing mechanism directs the request to a downstream process or sub-agent specifically trained or equipped to handle that particular intent. This modular approach is efficient and scalable.

A. A single retrieval pipeline is for information lookup (RAG) and inherently lacks the classification and routing logic needed.

C. Image segmentation is a computer vision task for partitioning a digital image and is irrelevant to text classification.

D. A speech synthesis engine converts text to audio; it does not perform any analysis, classification, or routing.

1. NVIDIA NeMo Framework Documentation, "Intent and Slot Classification". The documentation describes intent classification as a core task in Natural Language Understanding (NLU) for identifying the purpose of a user's utterance in conversational AI.

2. Jurafsky, D., & Martin, J. H. (2023). Speech and Language Processing (3rd ed. draft). Chapter 17, "Dialogue Systems and Chatbots". The chapter details task-oriented dialogue systems, which rely on NLU components to classify intent and extract slots to determine subsequent actions. (Source: web.stanford.edu/~jurafsky/slp3/)