Chapter 2: The knowledge base

The level 1 agent answers from your shop’s policies. For it to search them, the document has to be prepared first: read it, split it into fragments and store it in a knowledge base, where each fragment can be found by its meaning and not only by its exact words.

Create a blank project, for example Shop customer service. In this chapter you’ll build the part that loads the document, and in the next one, in the same project, the part that serves customers.

Note

Build both parts in the same project. The knowledge base that Chroma DB stores belongs to the flow where it is created.

The components

Component	What you’ll use it for
Drive File Manager	Reads the document from Google Drive.
Parse Data	Prepares the document’s text for splitting.
Split Text	Splits it into fragments.
Google Generative AI Embeddings	Computes the representation of each fragment that makes searching by meaning possible.
Chroma DB	Stores the fragments in the knowledge base.

Connect them

From	Output	To	Input
Drive File Manager	Data	Parse Data	Data
Parse Data	Processed Data	Split Text	Data Inputs
Split Text	Chunks	Chroma DB	Ingest Data
Google Generative AI Embeddings	Embeddings	Chroma DB	Embedding

Configure them

Drive File Manager

• Credential: your Google Drive credential.
• Operation: Get.
• Mode of Input: By Selection.
• Selection Mode: File.
• Select file: choose the policies document.
• In Advanced, turn on Process File Content. That way the component returns the document’s text, not just its details.

Parse Data

• Leave Template at {text}, its default value: it takes the text that Drive File Manager returns.

Split Text

• Chunk Size and Chunk Overlap come set to 1000 and 200. These values suit ordinary text documents: each fragment has about 1000 characters and shares 200 with the next one so an idea isn’t cut in half.

Google Generative AI Embeddings

• Credential: your Gemini credential.

Chroma DB

• Operation: Add.
• Collection Name: shop_policies. Note it down: the level 1 agent will search this collection.
• Leave Persist Directory empty.

Load the document

1. Click the play button on the Chroma DB component. This part of the flow runs up to that point.
2. When every component shows the green tick, the document is loaded.
3. Select Chroma DB and click Unplug in its toolbar. That way this part isn’t run again every time you test the flow.

If you change the document, undo the unplug and run it again.

In the next chapter you’ll build the part that serves customers.