Curious what others have experienced with this.

ODC's native Semantic Search is a solid starting point. You configure an entity attribute, pick a chunking method, and the platform handles the rest. For plain text content it works well.

What I wanted to understand was how it behaves with more structured content. Code blocks, tables, documents with heading hierarchies. So I built a small test app, loaded the same document into four entities each configured with a different chunking method, and ran the same query against all four. Then did it again with the same content as plain text.

The results were consistent across both rounds. The chunking methods are designed for flat prose. When the content gets more structured, the boundaries become harder to predict.

I wrote up the full findings with actual chunk output on Medium if it helps the discussion:  https://medium.com/itnext/what-odc-actually-does-when-it-chunks-your-text-8bf6b7b06a18

The test app is on Forge if you want to run it yourself: https://www.outsystems.com/forge/component-overview/24899/odc-native-chunking-tester-odc

Note that Semantic Search needs to be enabled on your environment to run it.

Has anyone else tested this with structured documents? Would love to hear what you found.