Embeddings
An embedding is a numeric vector that represents the meaning of text, an image, or other content in a mathematical space. Content that is close in meaning has close vectors, which makes it possible to find relevant documents even without exact keyword overlap, the foundation of semantic search and RAG.
In a RAG project, each documentation passage is turned into an embedding and stored in a vector database. When a user asks a question, that question is also embedded, then the nearest passages are retrieved. The quality of this link (embedding model choice, text chunking) often matters more for the final result than which LLM writes the answer.
A common trap is neglecting chunking and document quality while focusing only on the generative model. Embeddings over passages that are too long, too short, or poorly cleaned yield mediocre retrieval. You should also version the embedding model: changing it without reindexing the whole corpus silently breaks search relevance.
Follow Fragments Studio on Google
Add us to your preferred sources and our articles get surfaced first in Top Stories, AI Overviews and AI Mode.
