Embedding
An embedding is a text represented as a list of numbers that encodes its meaning. Two texts that say the same thing with different words produce similar lists, and that numeric closeness is what makes it possible to search by meaning rather than by exact match.
What is embedding?
An embedding is a text represented as a list of numbers that encodes its meaning. Two texts that say the same thing with different words produce similar lists, and that numeric closeness is what makes it possible to search by meaning rather than by exact match.
Also: vector · vector representation
Why it matters
What changes in a SaaS
It is the piece that lets a search find the right article even when the customer uses different words. Without embeddings, a help center only finds what is written with the same words as the title — which is almost never.
Embedding in detail
The intuition
Picture each text as a point in a space. Texts with similar meanings land near each other. Searching means turning the question into a point and looking at what documents surround it. The real dimensions number in the hundreds, but the idea holds.
How closeness is measured
With cosine similarity, which compares the direction of the vectors rather than their length. It is the usual measure because what matters is similarity of meaning, not how long the text is.
Passage size matters
Long passages dilute meaning and retrieve vaguely related things; short ones lose context and retrieve stray sentences. Tuning that size usually improves answers more than switching models.
They are recomputed when content changes
Update an article and its embedding has to be rebuilt, or search keeps finding the old version. It is the usual reason a system keeps answering with information that was already corrected.
Questions about embedding
What are embeddings used for in support?
To find the article that answers a question even when it is written with different words. They are the basis of meaning-based search in any modern help center.
Do embeddings need recomputing when documentation changes?
Yes. Otherwise search keeps finding the previous version of the text, and the system answers with information that was already fixed.
Related terms
A term on its own is only half understood. These come up in the same conversation.
RAG — Retrieval-augmented generation
Retrieval-augmented generation, or RAG, is the technique of finding relevant information in your own sources and handing it to a language model so it writes the answer with that. The model does not memorise the content: it looks it up at answer time.
LLM — Language model
A language model is a system trained on enormous amounts of text that, given a fragment, predicts how it continues. Writing, summarising, translating and holding a conversation all come out of that one simple capability: they are all ways of continuing a text plausibly.
Knowledge base
A knowledge base is the organised set of articles documenting how a product works and how to resolve what gets asked often. It serves two audiences at once: customers looking up the answer themselves, and the team using it as the source for their replies.
Help centre
A help centre is the public site where a company publishes its support documentation so customers can look things up themselves. It combines the articles with a search box, category navigation and, normally, the route to get in touch when the answer is not there.
Hallucination
A hallucination is an answer from a language model that sounds right and is not: an invented figure, a feature that does not exist, a plausible number nobody checked. It is not a bug, it is a direct consequence of how a model works — it produces the most probable continuation, not the verified one.
Tool calling
Tool calling is the mechanism by which a language model can run functions you define instead of only writing. You describe the available tools, the model decides which one to use and with what arguments, and it receives the result to fold into its answer.
