What is AI Builder π€
AI Builder is the platform for providing a low-code user experience for every developer to create and customize their PowerApps and Flows. You can access AI Builder from the navigation pane within the PowerApps Studio or the Microsoft Flow website.
The simple wizard-like experience is tailored to empower every developer in keeping with the essence of the Power Platform. You can access all your AI models in a single pane under the AI Builder βModelsβ tab, providing key information about each model upfront in a tabular format.
Prerequisites for using AI Builder in MS flow
To use AI builder in MS Flow, first, we need to create, train and publish an AI Builder Model. We can build multiple models for various scenarios like a From Processing Model to analyse and process your Invoices, an Object Detection Model to identify and count objects and so on.
Currently, to use AI Builder models in a flow, we need to create the flow inside a solution.
Use Predict Action to use AI Models
Predict Action takes Model and Request Payload as inputs and returns Response Payload as output. Predict action can work with different types of models, the output can contain different data.
We will use a Parse JSON action that will parse the output and make it available to subsequent steps with known names.
Click on the Content field and select Response Payload from the Dynamic content panel. Next, click on the Schema field and paste in the following. This is a schema that tells the parse action what to expect in the content.
We can find about difference ways to pass Request Payload and Schema to process the Response Payload in the Microsoft Docs site.
Now, once we get the result, it’s up to us to utilize it based on scenarios. Maybe we want to send an Email if we receive the invoice of an order greater than 1 Million dollars or maybe we should create a case if we receive negative feedback from our customer and assign it to an officer based on tags.
Opinion:
Although AI Builder is powerful enough to generate models and provide accurate outcomes, it all depends upon how we train it. We need to ensure that we provide sufficient, accurate and relevant data to ensure the best result out of it. We also need to keep records of failed scenarios, create a dataset of these record with correct values and keep training our model with a new version to improve the efficiency over time.
I hope I was able to persuade you to look into this feature and I hope it helps your client.
Till next time, Cheers π