UiPath Document Understanding - Generative Classifier and Extractor for Unstructured Documents

Generative AI is one of the trending topics today. It is a technology that helps generate new content, as well as process complex data and extract meaningful information. This also includes the processing of unstructured documents.

UiPath 2023.8 Public Preview now offers Generative Classification and Extraction for processing unstructured documents. These new models use generative AI capabilities, making them more efficient and flexible in processing unstructured content.

The video explains how we can use the new feature to process unstructured documents.

2 Likes

Have you tried this new feature with un-structured data @Lahiru.Fernando and if yes any insights from your side?

1 Like

Hi @Sandeep_Alexander_Goni

Yes, I have tried this on unstructured documents such as legal contracts. In my case we had to extract certain data points from the document and perform some transformations. The data points could be available in any page of doc because its a legal document. It was able to nicely extract the content that we looked for, while doing the little transformations.

Happy to help and guide in your use case if needed. Let me know…

Thanks
Lahiru

1 Like

Makes sense, thank you for the response @Lahiru.Fernando :+1: :slightly_smiling_face:

Hi, @Lahiru.Fernando . Do you know what’s working behind the scenes of the Generative Classifier and Generative Extractor activities? Is it a GPT or something else?

Hi @Joe_Matuch , currently they are experimenting with different models. The one used for now is GPT 3.5 Turbo (I think) which is hosted in an own environment and also finetuned for working with the activities.

1 Like

Hi @Lahiru.Fernando ,
I used Generative Extraction steps to extract my unstructured handwritten document data. The results are awesome. It was able to able digitize the correct values based on the prompts i gave under Generative Extractor action.

But here is my problem:
If i use Create Validation Task and Wait and passing the output of Generative Extractor, its throwing an error as variable types mismatch .
So

  • How to validate this data using Action center?
  • How bring Human in Loop to validate the data (Wait & Resume) Use case steps?
  • Where can i find the Generative Validation Activities?