Salesforce Docs Portal

Verify Custom Chunking Results

After your search index has run and reached a Ready status, verify that your custom chunking function produced the expected chunks. Inspect the chunk data in Data Explorer and, optionally, validate search behavior by using a retriever and Prompt Builder.

Edition Table
Available in: Developer, Enterprise, Performance, and Unlimited Editions. See Data 360 edition availability.
Permission Sets Needed
To verify custom chunking results in Data 360: Permission set: <ul><li>Data Cloud Architect</li></ul>

Prerequisites

  • A search index that uses your custom chunking function and has run successfully, meaning Search Index Last Run Status is Ready.

tip: Chunking runs only on content that was successfully ingested into the source you selected for the search index. If you expected chunks from a document, for example, a PDF, but see none, verify that the document uploaded and ingested successfully into that data source before the search index ran. A failed or incomplete upload means there is no content for the function to chunk.

View Chunked Data in Data Explorer

  1. Open the search index record and note the Chunk Data Model Object name listed on the page.

  2. In Data Explorer, select the appropriate data space, object type as Data Model Object, and the data model object name that corresponds to the chunk data model object from your search index.

  3. Open a row to view the chunk data and confirm that text, sequence numbers, and citations match what your custom function is designed to produce. Check that each chunk retains the context you intended (for example, a section heading with its content, or a product name with its description). Contextually coherent chunks lead to more accurate search and AI retrieval.

<!-- Commented out for a future release. Retain this section. ## Verify Search Behavior with a Retriever To confirm that search returns relevant chunks produced by your function: 1. Navigate to AI Models and go to the **Retrievers** tab. 2. Click **New Retriever**, select **Individual Retriever** as the retriever type, and click **Next**. 3. Select **Data Cloud** as the data source for the retriever. 4. Select the data space in Data 360 where the search index resides. 5. Select the data model object and the search index that uses your custom chunking function. Click **Next**. 6. Configure result count and fields, for example, chunk text and any citation fields. Click **Next**, review, and save the retriever. 7. Activate the retriever (for example, click **Activate** and confirm in the modal). 8. From the retriever list, open the dropdown for your retriever and select **Open In Prompt Builder**. 9. In Prompt Builder, create or use a prompt template, add a Search resource, and configure it to use your retriever. Map the search text input to the appropriate field (for example, an account or document field). 10. Enter a test question that should be answered by content you chunked, for example, a question about text that appears in your chunked documents, and run **Save & Preview**. 11. Review the response and citations to confirm that the retriever is returning chunks produced by your custom function and that search behavior meets your expectations. -->

See Also