This article outlines a technique for improving LLM performance in specialized organizational domains by using LLMs to generate context-specific glossaries. It emphasizes the importance of mapping bounded contexts to avoid confusion between different usages of terminology and recommends incorporating human expert review to ensure accuracy. The approach promotes using these glossaries to help agents navigate internal language, deprecated terms, and domain-specific nuances.
Key points
Use LLMs to extract domain-specific vocabulary and definitions from internal organizational text sources.
Human experts must review machine-generated glossaries to ensure technical accuracy and context correctness.
Maintain multiple, context-specific glossaries based on bounded contexts rather than a single, universal enterprise dictionary.
Glossaries should include guidance on term usage, including deprecated terms and preferred alternatives.
Version control and scheduled updates are essential for keeping glossaries relevant as organizational language evolves.