How a source becomes searchable
1
Read

An indexing job reads each document through the source's API, with the credentials you gave.

and more
Connect to the VPNConfluence
2
Split into passages

Documentation and articles: one sentence per passage, which keeps the sentences around it. Slack threads and Jira issues: larger chunks.

Open the VPN app from Self Service.
Enter vpn.northwind.com as the portal.
Sign in with your Northwind email and approve the push.
On hotel Wi-Fi, switch to the TCP profile.
passagecontext kept with it
3
Compute embeddings

An embedding model on AWS Bedrock turns each passage into a list of numbers that stands for its meaning.

"Sign in with your Northwind email and approve the push."
[0.12, −0.48, 0.07, …]
1,024 numbers per passage
4
Store

Each passage goes into your organization's own table in the search database: its text for keyword search, its embedding for semantic search, and its labels.

Northwind's table
Text and embeddingLabels
Sign in with your Northwind…
[0.12, −0.48, …]
ConfluenceITEveryone