High Performance PostgreSQL for Rails

Storing and Searching Vector Embeddings

Large Language Models (LLM) developed by companies like OpenAI have grown tremendously in 2023.

API services accept text input and produce “embeddings” from the text. Embeddings are “numerical representations of text.” These numerical representations are then stored in PostgreSQL and can be indexed for high-performance search queries. The advantage of this approach, just like with tsvector objects, is that the processed search objects are stored right alongside your relational data.

What’s the flow like for this integration? The API service is said to embed the text, and then returns a vector (or list) data structure. Vectors are lists like the Array type in Ruby. The vector has one or more floating ...