Question

A pharmaceutical company is building a custom agent in Microsoft Foundry to answer regulatory affairs questions. The base foundation model performs well on general reasoning but frequently misinterprets the company's highly specialized internal drug-naming conventions and abbreviations, producing incorrect terminology in its responses. The company has a large, well-labeled corpus of internal documents that consistently use these conventions. Latency and answer freshness are not primary concerns, but the responses must reliably adopt the company's linguistic style and terminology. Which approach should the architect recommend to best address the terminology problem?

Accepted Answer

Fine-tune a custom model in Foundry using the labeled internal corpus so the model internalizes the company's terminology and style

Answer

Configure retrieval-augmented generation (RAG) grounding over the document corpus without modifying the base model

Answer

Add a longer system prompt listing every drug abbreviation and naming rule to steer the base model at inference time

Answer

Switch to a smaller language model to reduce the chance of the model inventing incorrect terminology