Knowing that you have to do that as a separate step adds a whole additional level of complexity too. For example, if some content has the images and some don't

Knowing that you have to do that as a separate step adds a whole additional level of complexity too. For example, if some content has the images and some don't, you need to add whole additional steps to your processing and potentially add hallucinations in. What are you using for document extraction lately, Simon?