NLP Fundamentals
Vector Database
A specialised database optimised for storing high-dimensional vectors and retrieving nearest neighbours efficiently.
Definition
A vector database is built around fast approximate nearest-neighbour search on high-dimensional vectors — most commonly the embeddings produced by language models. Where a traditional database asks 'which row has exactly this id?', a vector database asks 'which 50 vectors lie closest to this query vector?' That capability underlies most modern retrieval-augmented AI: a user question is embedded, the database returns the most relevant snippets, and the language model writes an answer grounded in them.
Why it matters
Vector search makes it practical to ground AI output in your own content — product documentation, knowledge base, prior support tickets, internal policy — without retraining a model. That shift is one reason AI products improved so quickly between 2023 and 2025: instead of expensive fine-tuning, teams point the model at the right context at query time. For content teams, the implication is straightforward. The quality of your written corpus increasingly determines the quality of the AI experience layered on top. Clear, complete, well-structured writing makes retrieval-augmented systems work; vague or contradictory writing degrades them.
Examples
Support automation
Embedding 4,000 historical support articles into a vector database lets an assistant pull the three most relevant on each customer message, grounding the reply.
Internal search
An employee types 'how do we handle expense reports over $500?' and the system surfaces the right policy paragraph even if the doc title never mentioned dollars.
Frequently asked
Do I always need a dedicated vector database?
Not for small corpora. Up to a few thousand vectors, an array and a similarity function suffice. Beyond that, purpose-built tools become worth the operational cost.
What about Postgres with pgvector?
A common starting point: it adds vector capabilities to a relational database many teams already run. Performance is good into the low millions of vectors.
Does this replace search engines like Elastic?
Often complements rather than replaces. Hybrid search — keyword plus vector — typically beats either alone.
Related terms
Embeddings
Numerical vector representations of text whose distances capture semantic similarity between pieces of content.
Retrieval-Augmented Generation (RAG)
A pattern in which a model is given relevant context retrieved from your own sources before generating an answer.
Semantic Search
Search that retrieves results by meaning rather than exact keyword match, typically using embeddings and vector similarity.
Put the concept to work
Open the rewrite engine and apply this principle to a draft of your own.
Try a rewrite