Vector Database

You search for "affordable places to live with good weather" on a traditional search engine. It looks for documents containing those exact words. "affordable," "places," "live," "good," "weather." If a blog post about "budget-friendly cities with year-round sunshine" doesn't contain your exact keywords, you'll never find it. Same meaning, different words.

Now imagine a search system that understands meaning. You type your query, the system converts it into a mathematical representation of its meaning, a vector of 1,536 numbers, and finds the documents whose vectors are closest to yours in meaning-space. The blog about "budget-friendly cities with sunshine" is mathematically close to "affordable places with good weather," even though they share almost no words.

This is what vector databases enable. They store and search embeddings, high-dimensional vectors that capture the semantic meaning of text, images, audio, or any other data. An embedding model (like OpenAI's text-embedding-3-small or Sentence Transformers) converts data into these vectors, and the vector database finds the most similar ones.

Vector databases are the infrastructure behind:

• RAG (Retrieval-Augmented Generation). ChatGPT-style AI that retrieves relevant documents before generating answers
• Semantic search. Finding content by meaning, not keywords
• Recommendation engines. "Users who liked this also liked..."
• Image similarity. "Find me photos similar to this one"
• Anomaly detection. "This transaction's vector is far from normal patterns"

A vector is just a list of numbers: [0.12, -0.45, 0.89, 0.03, ...]. In machine learning, these numbers represent features of the data, dimensions of meaning.

An embedding is a vector produced by a trained model that captures the semantic essence of the input.

"The cat sat on the mat" → [0.12, -0.34, 0.56, 0.78, ..., 0.91] (1,536 dimensions)
"A kitten was resting on a rug" → [0.11, -0.33, 0.55, 0.79, ..., 0.90] (very similar!)
"Stock prices fell today" → [0.87, 0.23, -0.91, 0.04, ..., -0.67] (very different)

Notice: the cat sentences produce nearly identical vectors because they mean nearly the same thing. The stock market sentence produces a completely different vector because the meaning is unrelated.

Similarity is measured by distance. The two most common metrics:

Cosine similarity: Measures the angle between two vectors. Values range from -1 (opposite meaning) to 1 (identical meaning). This is the most common metric for text embeddings because it ignores magnitude and focuses on direction.

Euclidean distance (L2): Measures the straight-line distance between two points in vector space. Smaller = more similar.

The core operation of a vector database: given a query vector, find the K vectors in the database closest to it. This is the K-nearest neighbors (KNN) problem.