Vector Databases

Chapter 2. Embeddings

At their core, vector embeddings solve a fundamental challenge in computing: how to represent complex, unstructured data in a way that machines can understand and efficiently process. In this chapter, you’ll dive deeper into the history, evolution, and current usage of vector embeddings, with code examples showing their applications in language models. This is a conceptually meaty topic, and you may find yourself revisiting it more than once when faced with unfamiliar terminology or concepts in later chapters as you begin applying embeddings.

Understanding Vector Embeddings: Why We Need Them

Traditional data structures work well for structured information such as numbers, dates, and categorical values, but they fall short when dealing with the rich, nuanced content humans naturally create and consume—text, images, audio, and video. Embeddings bridge this gap by mapping raw data into vectors (ordered lists of numbers) that capture meaning and relationships within the data (see Figure 2-1).

This transformation from unstructured data into mathematical vectors isn’t just a technical convenience—it’s the foundation that enables modern AI systems, including LLMs and generative AI, to process and understand information in ways that approach human comprehension. Unlike traditional feature vectors based on hand-crafted statistics, embeddings are learned representations in which proximity in vector space reflects semantic or conceptual similarity. This mapping from ...