A vector is a list of numbers that represents the meaning or features of something as a point in space. By turning a sentence, an image or a product into a vector, software can compare two items by how close their points are, rather than by matching exact words.
A simple analogy is placing words on a map. On this map, “dog” and “puppy” land near each other, while “spreadsheet” sits far away, even though none of them share letters. The closeness on the map is the similarity in meaning. These vectors are produced by machine learning models, and an LLM uses the same idea to understand how concepts relate. When you have thousands of them, you store them in a vector database built to find the nearest matches quickly.
In practice a real vector is not two numbers on a flat map but a long list, often hundreds or even thousands of values. Each value captures one fine-grained aspect of meaning, and together they place an item very precisely in a space no human could draw. Similarity is then just a distance calculation between two of those points.
This is what makes modern semantic search and recommendations work. Ask for “warm winter jacket” and a vector search can surface a coat described as “insulated parka for cold weather,” because their vectors are close even though the words differ.
One caveat: a vector only reflects whatever the model learned. Feed it niche jargon or a brand-new product it has never seen, and the placement can be off, which is why teams test results on real queries rather than trusting the math blindly.
At TopDevs we use vectors to build search and AI features that understand intent, so a client’s users find the right thing even when they do not type the exact keyword.
