Skip to main content

Chunk Management

When a file is ingested into a Knowledge Base, SimplAI parses and splits its content into chunks based on your configured chunking strategy. These chunks are then embedded and stored for retrieval. Chunk Management lets you review and adjust those individual chunks after ingestion, so you can correct parsing artifacts, remove noise, and refine content without re-uploading or re-processing the entire source file.

Why Edit Chunks?​

Automated parsing and splitting work well for most content, but some documents produce chunks that are not ideal for retrieval. Editing chunks directly allows you to:

  • Fix parsing artifacts: Clean up broken text, stray characters, or formatting issues introduced during extraction.
  • Improve retrieval accuracy: Refine chunk text so that the most relevant content is matched against user queries.
  • Remove noise: Drop chunks that contain irrelevant content such as headers, footers, or boilerplate.
  • Add missing context: Adjust chunk content so it stands on its own when retrieved in isolation.

Viewing Chunks​

  1. Navigate to the Knowledge Base section in SimplAI and open the Knowledge Base you want to work with.
  2. Select the ingested file whose chunks you want to review.
  3. SimplAI displays the list of chunks generated from that file, in document order.

Each chunk shows its extracted text so you can assess how the source content was divided.

Editing a Chunk​

  1. Locate the chunk you want to change in the chunk list.
  2. Open the chunk to enter edit mode.
  3. Update the chunk text as needed.
  4. Save your changes.

When a chunk is saved, SimplAI re-embeds the updated content so the change is reflected in subsequent retrievals.

Refining and Validating Chunks​

After editing chunks, use Retrieval Testing to confirm that your changes improve results. Run representative queries and review which chunks are returned to verify that the edited content is being matched as expected. Iterate on chunk edits and your chunking strategy until retrieval quality meets your needs.

Best Practices​

  • Edit selectively: Focus on chunks that are noisy, incomplete, or frequently mis-retrieved rather than reviewing every chunk.
  • Preserve context: Keep enough surrounding context in each chunk so it remains meaningful when retrieved on its own.
  • Validate with retrieval testing: Confirm the impact of edits with real queries before relying on the Knowledge Base in production.
  • Revisit your chunking strategy: If many chunks need editing, the underlying chunking configuration may need adjustment instead.