Check Vector Store and Data Synchronization

Target

Before the AI can answer, data must be ingested into the vector storage (Vector Store). We will perform a “Before and After” check to clearly see how Bedrock automatically encodes and stores data into OpenSearch.

Implementation Steps

Step 1: Check Vector Store (Empty State)

We will directly access Amazon OpenSearch Serverless to confirm that no data exists yet.

1. In the AWS Console search bar, type Amazon OpenSearch Service and select Amazon OpenSearch Service.

2. In the left menu, under Serverless, select Collections.

3. Click on the Collection name newly created by Bedrock (usually named like bedrock-knowledge-data...).

4. On the Collection details page, click the Open Dashboard button (located at the top right of the screen).

• Note: If asked to log in, use your current AWS credentials.

5. In the OpenSearch Dashboard interface:

• Click the Menu (3 horizontal lines) icon in the top left corner.

• Select Dev Tools (usually located at the bottom of the menu list).

6. In the Console pane (on the left), enter the following command to check data: GET _search { "query": { "match_all": {} } }

7. Click the Play (Run) button (small triangle next to the command line).

8. Result: Observe the right pane, hits -> total -> value is 0.

Step 2: Sync Data

Now we will trigger Bedrock to read files from S3 and load them into OpenSearch.

1. Return to the Amazon Bedrock tab on the browser.
2. Select Knowledge bases in the left menu and click on the KB name you just created.
3. Scroll down to the Data source section, check the box (tick) next to the data source name (s3-datasource).
4. Click the Sync button (Orange).

5. Wait:
   • This process will take 5 - 10 minutes depending on the sample document size.
   • Wait until the Sync status column changes from Syncing to Available.

Step 3: Re-check Vector Store (Populated)

After Bedrock reports Sync completion, we return to the repository to verify the data has been successfully ingested.

1. Switch to the OpenSearch Dashboard tab (still open from Step 1).
2. In Dev Tools, click the Play (Run) button again with the old command:

   GET _search
   {
     "query": {
       "match_all": {}
     }
   }
   

3. Result:
   • The hits -> total -> value section will be greater than 0 (e.g., 10, 20… depending on the number of text chunks).
   • You will see details of the vectors (number arrays) and text content stored in the _source field.

Congratulations! You have completed building the “brain” for the AI. The data has been encoded and sits safely in the Vector Database, ready for retrieval.