Knowledge engineering as a function

None of the above happens by accident, and it does not happen as a side project of data engineering or of the AI team. It requires people whose job is the knowledge base: designing chunking and contextualization strategies for each content class, managing taxonomies and ontologies, tuning extraction, building and refreshing evaluation sets, measuring retrieval quality, operating the catalog, running contradiction resolution, and supporting domain stewards through onboarding.

In most organizations this function does not exist yet. Its absence is the reason knowledge bases decay. The people who build the pipeline move on to the next project; the people who own the content do not know the pipeline exists; nobody is watching the freshness metrics; and a year later the copilot is answering from last year's documents and nobody can say why.

The function is small — a few people for a platform serving many domains — and it sits between data engineering and the domains, with a clear mandate: the knowledge base is an asset, and it is measured, current, and trusted, or it is not a knowledge base.