article --summary
The summary
The post explains why large organizations struggle with specialized vocabulary when using LLMs, since internal terms are absent from training data. It proposes having an LLM scan internal documents, emails, and chat logs to draft a dictionary-style glossary that includes both new terms and familiar words used in unusual ways.
The draft should then be checked by domain experts, stored where other models can easily consult it, and refreshed on a schedule under version control. The author advises keeping separate glossaries for each bounded context rather than one standard glossary, since the value often lies in showing how the same term differs across contexts.
Glossaries can also guide code generation, flag deprecated or discouraged terms, and keep internal jargon out of customer-facing interfaces.
Editorial note: This is a summary of an article first published by Martin Fowler. The full piece is on their site.