Skip to main content

Unlock Document Parsing with Cradl AI in UiPath


This guide will show you how to extract data from your documents with Cradl AI in UiPath. By the end of this guide you should be able to:

  • Train your own AI model customized to your own documents.
  • Set up a feedback loop so your AI model can train and get better for every document you process.
  • Process your documents in your UiPath workflow using Cradl AI.

All you need to get started is:

  • UiPath Studio installed on your computer.
  • Account on Cradl AI (If you don't have one yet, you can sign up here).
Scalable

Cradl AI empowers you with the ability to create your own AI models. 15 examples are all you need to get started.


Overview


The Cradl AI integration for UiPath can be downloaded here, and consists of these activities:

NameDescription
Parse DocumentParse a document with a model in Cradl AI.
Parse Document With Human In The LoopParse a document with a flow in Cradl AI.
List Processed DocumentsRetrieve data from a flow in Cradl AI.
Mark Run As CompletedSet a run's status to "completed"

Step 1: Login to Cradl and setup a Flow


There is very little setup required in Cradl before you are ready to use the activities in your UiPath workflows. All you have to do is:

  • Open the Flow Builder and press create new.
  • Select a pre-trained model to get started. You are also free to customize the model to extract the fields you want.
  • Add your co-workers as human validators to get help with the initial validation process.
  • Give your flow a suitable name and press save.

You are now ready to jump into UiPath Studio and start the integration. But on your way there, remember to bring the parameters listed under UiPath in the input and/or export dropdown. They are essential to identify you, your model and your flow. Here is a brief explanation of what they are and how they are used:

ParameterTypeDescription
clientIdAuthenticationThe identity of your UiPath integration. Store it as a variable in the Data Manager.
clientSecretAuthenticationThe secret key to authenticate your client. Do not share this with anyone, but store it safely as a variable in the Data Manager.
flowIdInputThe identity of the flow you just created. Store it as a variable in the Data Manager.
Adaptive

You can at any point change the model in your flow by selecting another modelId.


Step 2: Cradl AI Example Project in UiPath


  • Download the Example Project.
  • Unzip the folder and open the project in UiPath Studio.
  • Open the Data Manager in the side panel and store the parameters from Flow Builder.

Step 3: Send a document to your flow in Cradl AI and complete a step of the feedback loop


  • To test that everything works, open Start Execution.xaml from the zipped folder and do a test run. Make sure that clientId, clientSecret, and flowId all have correct values.
  • Head over to Cradl. You can now go to the Validator by pressing the button Run In Validator. This is the Human-in-the-loop platform that comes with Cradl AI, and you should see your document appear in your flow.
  • To get your own customized model you need to provide some ground truth data, so the AI knows what to look for. This is done by heading down to Tasks.
  • Fill in the correct values and press the big green button. You have now completed one iteration of the feedback loop.

The values that you annotated are used as ground truth when training the model, so you are one document closer to having your own customized model!

tip

You need to have at least 15 documents annotated this way before training a model.

So just to sum up, we have now done the following:

  • Sent in a document for parsing to Cradl AI
  • Created a feedback loop and done one iteration of it.

Until now this seems like just another manual step in your workflow, where is the automation?
The automation starts after 15 documents have passed through the manual step, then you can train your model. When the training has completed, adjust the confidence thresholds in your flow so correctly predicted values from the model pass straight through the validation. This brings us to the next section, how do we get the predictions from the model back into UiPath?


Step 4: Get the parsed document back into UiPath


  • Open the sequence defined in Parse Document From Flow.xaml.
  • Make sure that clientId, clientSecret, and flowId all have correct values. Then do a test run.
  • A message box will pop up for each processed document showing the output from Cradl AI. If not, check that you have processed at least one document in Cradl. We are only listing succeeded runs here, so all runs that are waiting in line to be parsed manually will not show up here.
  • The demo process will loop through all the succeeded runs and mark them as completed, so you don't risk sending them further down your pipeline several times.

In this example, we simply print the data in a message box. This is just to illustrate how easy it is to retrieve the output from Cradl and pass it along to other stages in your workflow. The rest is up to you, you can write the data from each run to your database, send it to another API, or just safely store it in a local file. Happy document parsing!