Question

A financial services firm wants an agent to classify inbound support emails into 12 predefined categories and route them. The workload is high-volume, the classification taxonomy is narrow and stable, responses must be near-instant, and the compliance team requires that no email content leave the firm's own Azure tenant boundary. As the solutions architect, which model strategy best fits these requirements?

Accepted Answer

Fine-tune a small language model (SLM) hosted within the firm's Azure environment and use it for the classification task

Answer

Call a large frontier foundation model through a public API endpoint for every email to maximize classification accuracy

Answer

Build a fully custom transformer model from scratch trained on the firm's entire historical email corpus

Answer

Deploy a multi-agent orchestration solution where several large models vote on each category assignment