Why it matters
Embeddings translate text into vectors that capture semantic relationships. Poor embeddings mean semantically related policy and FAQ never neighbor each other in search.
How it works
Choose models suited to your domain and languages. Re-embed when models change or when documents shift materially. Version embedding configs alongside index builds.
Example
Nintendo regenerates embeddings for Refund Policy when section headings change materially, triggering regression tests because similarity neighborhoods shift.
Common mistakes
- 1Embedding titles without body text for long articles
- 2Mixing embedding model versions in one index
- 3No spot checks after re-embed jobs
Definitions are useful. Governed knowledge is better.
Wiki helps you turn these concepts into a real system for the knowledge behind your AI.
See Wiki in action