Building a RAG implementation piece by piece

In Chapter 17, we implemented a RAG solution using Vertex AI Search. I explained that Vertex AI Search makes the process very easy for us because it abstracts away the steps in the process, such as chunking and embedding our content, and it performs all of those steps for us behind the scenes. There are also popular frameworks, such as LlamaIndex, that help to simplify RAG implementations, and Vertex AI has also launched a grounding service (Vertex AI Grounding) that you can use to ground responses from a generative model with results from Google Search or with your own data (using the aforementioned Vertex AI Search solution). In this section, we will dive deeper into the process and build a RAG solution, piece by piece. Before we dive into the solution architecture, we’ll cover some of the concepts that had been abstracted away in Chapter 17, most notably regarding tokens and chunks.

Tokens and chunks

In this section, we&...