UST SmartOps™ uses cognitive automation to intelligently and continuously accelerate business opportunities by reducing costs, streamlining efficient operations, and uncovering unrealised growth revenues for your business.

SmartVision Release 2.3

Release Date: 26 May 2021

Intuitive trainer module to support SmartExtract model training through document annotation.

Capability to host local & global (multiple) models

Field enrichment/Auto-fill provision in document review screen

Overview

SmartVision 2.3 introduces the Document Understanding Trainer module that builds SmartExtract machine learning models through document annotation. It follows a no code approach and doesn’t require any technical knowledge for training. We envision an accelerated onboarding of new document types & FOI expansion leveraging this. To cater to the varied document types we are enhancing DU to support local & global (multiple) extraction models. As the first step towards multi-lingual enablement, US Invoice SmartExtract model is enhanced to support data extraction from English & Mexican Spanish invoices.

SmartVision PWF now includes the field enrichment/auto-fill provision which allows you to populate dependent fields from a master lookup file or data dump. It also includes enhancements to point & select, table rebounding features, ability to configure duplicate invoice validation and changes for dynamic URL.

Document Understanding Trainer (DU Trainer)

Introducing DU Trainer module, that supports SmartExtract model training based on document annotation. User has the flexibility to create a training project, upload the training document sets and annotate the required fields of interest(FOIs) for both headers & tables. On completing the annotation process, model training can be triggered. Trainer includes an intuitive UI for document annotation, enriched with “point and select” & “table bounding” tools which makes the document annotations fast & efficient.

On an average we expect around 100 representative samples for training purpose. Training can be done for any forms type of documents like invoices, purchase orders, remittance advices etc.

DU Trainer follows a no code approach and the primary objective is to accelerate the onboarding of new document types & additional FOIs.

How to set up a DU Training project?

Key Features - DU Training project creation, Upload Training documents, Document Search & alternate flows

How to annotate Fields of Interest in documents?

Key Features – Insights & Table FOI annotation

How to initiate model training?

Key Features – Annotation Summary View, Initiate training

How to predict using the new model?

Key Features – Publish models to DU

Capability to host local & global (multiple) models

 
  • Currently DU can host only one global model for an extraction engine. However with SmartExtract, we expect more models to be created for each document type. Hence, we are adding the capability to support multiple SE models globally. Global SE models would reside under locale/domain/document type umbrella. For e.g. invoice models for US & AU locale can be hosted separately under US/Finance/Invoice & AU/Finance/Invoice respectively. A project created for US/Finance/Invoice would automatically pick the invoice model configured for US locale.

    In addition, we are also adding the capability to override the global models from project. User can attach a model to the project via 2 options.

     

    1. Publish the model from DU Trainer by providing security token (applicable when DU & DU Trainer resides in the same environment)
    2. Import the configurations from DU Trainer & manually integrate the model (applicable when DU & DU Trainer resides in different environments)

     

Pre-trained Models

 
  • Invoice model for Australia locale [AUS/Finance/Invoice]
  • Claims adjustment note model for Australia locale [AUS/Finance/Adjustment Notes]
  • US Invoice model enhanced to support Mexican Spanish extraction along with English extraction.

Field enrichment/Auto-fill provision in document review screen

 
  • Field enrichment/Auto-fill capability in document review screen provides the provision to populate dependent fields from a master lookup file or data dump. The “Auto-fill” action invokes an external custom API from the review screen and allows the customer specific business rules to be handled at the custom layer. Both insights & table line item overrides are possible with this option. Examples as below:

     

    1. Vendor details like vendor name, vendor address etc. can be populated by passing vendor code from review screen.
    2. Line items from a purchase order can be fetched by providing purchase order from review screen.

     

Feature enhancements for HITL corrections

 
  • Extending the Point & Select feature for table line items
  • Improvements in Table column rebounding feature
    1. Drawing pane reset option
    2. During rebound, replace the existing rows with the newly bound rows (at the same location) and highlight for easy identification
    3. Enhance the ability to adjust the table boundaries.

Configuration for duplicate invoice validation

 

As part of the project creation, user will have the provision to configure duplicate invoice validation. Duplicate validation based on Invoice number & Vendor Code will be applied only if it is configured in the project.

Dynamic URL changes

 

SmartVision URL is upgraded to dynamic URL. paas/invoiceextraction in the current URL shall be replaced to be configured dynamically.

Accuracy calculation updates in Analytics dashboard

 

If the user entered data matches with the system extracted data (default & candidates) in Insights section, then it would be counted as accurate for accuracy calculations. This would help us eliminate the overrides by user in spite of the extracted data being accurate.

  • (SDI-330) Navistar - Admin was able to assign invoice to deleted user.
  • (SDI-773) Navistar: Performance Issue -7.1
  • (SDI-797) Navistar - Invoice went missing after extraction
  • (SDI-956) Navistar - DB credentials showing as plain text in container log
  • (SDI-969) Navistar - Smartvision Icon is not properly aligned in header section.
  • (SDI-976) Navistar - Blank space is appended with extracted field after delete existing data in INSIGHTS section
  • (SDI-1026) Navistar - Split invoice status field updated as Unassigned
  • (SDI-1028) Navistar - Showing different behavior line item field validation
  • (SDI-1029) Navistar: Some field value was turned to null when user tried to edit
  • (SDI-1030) Navistar: Extra space is added automatically during the editing textbox
  • (SDI-1031) Navistar: Error on Invoice Number field was retained after successful approval
  • (SDI-1033) Navistar - Future date is updated on date picker for Advanced Filter

Release Artifacts

Deployment Artifacts

Question

How many global model(s) DU can host for an extraction engine ?
 
Send your answer and get exciting Prizes

First one who send the correct answer will grab the prize.
Now what are you waiting for...

Winner of last Quiz

John Doe (U32057)

Question:
User can attach a model to the project via 2 options. Is this statement Correct or Wrong?
Answer:
Correct

For Support and more enquiries you can write a mail to us on feedback.smartops@ust-global.com