Vector Databases

Though the previous databases have different storage models, they share a common strength. They store facts - names, dates, relationships, etc.

But what about ideas? What about the abstract relationships that float around data?

Let’s think about our Pet data again.

What if we want to search the database for “dogs that need a lot of space”?

In our database, we might have a field for “size” - which could be related to space. We may have one for “energy” - which may mean a dog needs space to run and play. But neither of those fields match the keyword “space”. They don’t match the abstract idea behind needing space.

So our traditional databases would struggle with that kind of search.

That’s where Vector Databases come in. (And why LLMs use them).

Vector databases turn information into a “vector” and store it in a multi-dimensional space.

What does that mean?!

Let’s break it down starting with what a vector is.