This video explains how vector databases speed up similarity search, and why traditional indexing does not work.
In relational or NoSQL databases, indexing a column allows fast lookups using binary search.
That works because the data is one-dimensional or sorted on a single column.
In contrast, vector databases deal with multiple dimensions.
Sorting across all dimensions is not possible.
Binary search cannot be applied in a high-dimensional vector space.
Instead, vector databases use specialized techniques to find approximate nearest neighbors.
This includes building spatial partitioning structures like HNSW or IVF.
The key point is that vector search is fundamentally different from regular indexed search, and it requires its own optimized approach.