Technical Architecture of Vector Databases
Now one would think how a Vector database actually works. Let's dissect the technical architecture of vector databases to unravel their inner workings. At a fundamental level, they comprise two key components:
- Storage layer: This layer is responsible for efficiently storing and indexing the high-dimensional vectors. It leverages specialized data structures such as inverted indexes, tree-based structures (e.g., KD-trees), or state-of-the-art techniques like Product Quantization to optimise storage and retrieval operations.