Document Understanding AI - 5 new models in Preview (Cloud and On prem)

New ML Models in Public preview!

UiPath Document Understanding is launching 5 more ML Models in Preview on both Cloud and On Premises deployments. All models are retrainable (except for OCR engine), so their objective is two-fold. First, you get new, Out-Of-the-Box, Document Understanding capability. But perhaps more importantly, you get a leg-up for training a custom ML model for your specific usecase, by using these models as Base Models in AI Fabric (both On Premises and in the Cloud).

For instance in the On Premises version, when creating a new ML Package you will now have these options. This will make it an order of magnitude easier to train your own Indian Invoices model (for instance) since our OOB Preview model already gets you 90% of the way there.

  • OCR - UiPath proprietary OCR engine covering Latin script text. To be called from UiPath Document OCR activity in OCR.Activities Pack.

  • Purchase Orders - to be called from ML Extractor activity released in 2019.6. List of fields recognized:
    PO number
    Client Name
    Client Address
    Vendor Name
    Vendor Address
    Shipping Name
    Shipping Address
    Billing Name
    Billing Address
    Payment Terms
    Delivery By Date
    Net Amount
    Tax Amount
    Tax Rate
    Total Amount
    Line items: Line Number, Description, Product Code, Delivery Date, Unit of Measure, Unit Price, Quantity, Line Net Amount, Line Tax Rate, Line Tax Amount, Line amount
    For the always up to date list of fields see here, and make sure to have a JSON viewing extension installed on your Web Browser.

  • Utility Bills - to be called from ML Extractor activity released in 2019.6
    Billing Name
    Billing Address
    Vendor Name
    Vendor Address
    Account Number
    Invoice Number
    Total Amount
    Invoice Date
    Due Date
    Payment Address
    Service From Date
    Service To Date
    Payment Terms
    Previous Balance
    Document Type (Invoice or Statement)
    For the always up to date list of fields see here, and make sure to have a JSON viewing extension installed on your Web Browser.

  • Invoices-India - invoices extraction model including 9 additional fields, specific to the India market, to be called from ML Extractor activity.
    Supplier GSTIN
    Vendor GSTIN
    SGST Percentage
    CGST Percentage
    SGST Total
    CGST Total
    IGST Percentage
    IGST Total
    CGST SGST Total
    For the always up to date list of fields see here, and make sure to have a JSON viewing extension installed on your Web Browser.

  • Invoices-Australia - invoices extraction model including 2 additional fields, specific to the Australia market, to be called from ML Extractor activity.
    For the always up to date list of fields see here, and make sure to have a JSON viewing extension installed on your Web Browser.

Distribution channels :

Document Understanding on premises.
Public Endpoints.
AI Fabric Cloud.

Please don’t forget to send us your feedback so we can improve these preview features and make them shine in your workflows!

The Document Understanding Team


well this is going to be fun :slight_smile:


Are there any plans to add bank statements as well?

Hello @botBotGo,

We are interested in adding bank statements as well, but we have not yet obtained enough samples to train an ML model. We would need around 1000 diverse samples for that.


1 Like

I want to understand if Document Understating Framework has any relation to AI fabric or not? Is there any clarity on it ?

Hi @Gaurav_Sharma thank you for your question,

Document Understanding Framework is our suite of Activities, centered around the IntelligentOCR activity pack (and including some other packs like DocumentUnderstanding.ML and OCR.Activities), which enables document processing. This document processing involves AI capabilites and non-AI capabilities. Those capabilities which involve AI require special infrastructure, and that infrastructure is called AI Fabric. So basically the DU Framework relies on AI Fabric for those components which involve AI. Customers which don’t need AI features, won’t need AI Fabric. Customers which do need AI features, will require AI Fabric.

Does this help clarify things?

1 Like

Thanks. It does to certain extent. However where can I find more details or even an visual architecture of it. It is bit confusing. E.g in DUF there are different endpoints mentioned like , du.uipath com . What are these ? Is it reasonable to say that we would required AI fabric if ML extractors are used and not for non-ML extractors?

Yes, this understanding is correct.


Thanks. That clarifies it. Also, at the moment Uipath has some ML endpoints like that are consumed by ML extractors. Are these endpoints on AI Fabric? In essence, what I would like to have clarity is if a DUF installation can be independent of AiFabric or is it tightly coupled to it.

Hello @Gaurav_Sharma,

Some ML Endpoints are hosted, managed and trained by UiPath. You don’t need AI Fabric (Cloud/On-Prem) to access these end points. All you need is the following activity:


Now if you want to create your own ML Model & then use it to extract details from a document then you need to use AI Fabric (Cloud/On-Prem) to deploy and train ML Models.

Hope this helps ! Kudos :slight_smile:


Thanks . Yes it helps. To clarify it further, if I have my entire Uipath setup on 2019.x and on-premise (with orchestra or and robots VM as well on own data center), can I still use just the Document Understanding Framework from Uipath Automation cloud through enterprise license key and enterprise endpoints ?

Hello @Gaurav_Sharma,

Yes you can.


1 Like

Hello @alexcabuz,

I am getting below error while while getting the capabilities: UiPath.DocumentUnderstanding.ML.Activities.Exceptions.MLRequestException: Invalid server response. —> System.Net.Http.HttpRequestException: Response status code does not indicate success: 404 (Not Found).
at System.Net.Http.HttpResponseMessage.EnsureSuccessStatusCode()

Please assist as soon as possible.

Can you please assist

Can you please try with https?

1 Like

Please let me know if your issue is resolved

Yes Issue has been resolved by upgrading package version of ML extractor. Thank you for your help.


How do you train invoice models using cloud AI fabric ? I am not sure how to use data manager for cloud version.

Hi @Raja_Krovvidi please take a look at the following documentation pages:

If you follow these instructions and still have trouble, please let us know.


Is there any model to extract information from Tax Document ?