Create a grounded prompt in AI Builder - Dataverse(preview)- Use prompt in a cloud flow

-

Creating a grounded prompt in AI Builder using Dataverse is an innovative way to enhance AI capabilities by connecting them to a structured data source. Grounding AI prompts involves linking them to specific knowledge, in this case, the data stored within Dataverse. This process allows for more contextually relevant and accurate responses from AI systems, as they can draw directly from the data available in Dataverse.

You can use prompts in various scenarios, such as to:

  • Generate summaries or reports from your data.

  • Generate categories or reports from your data.

  • Answer questions from your customers or employees.

  • Extract named entities from text.

  • Draft a reply to a request.

AI Builder provides a prompt builder that allows makers to write, test, save, and use prompts to generate text.

Lets go and build a prompt now!

To create a prompt in AI Builder, follow these steps:

  1. Sign in to Power Apps or Power Automate.

  2. On the left pane, select the ellipsis (...) to expand the More menu and then select AI hub.

  3. Under Discover an AI capability, select AI prompts.

  4. Select Create text with GPT using a prompt.

  5. Give your prompt a name.



Here I am using the quote entity to create the prompt. 

On the right pane, select Input, select + Add input, and then name your input Quote Summary Prompt

On the right pane, select Data used (Preview). Select + Add data. Expand and select the Quote entity. select a data source from the Filter attribute dropdown menu. Enter a name in the Filter value field. Select + Add value and then select Quote Summary Report

Now write the prompt in natural language by inserting the required attributes from the Quote entity. We can also include the information from the related entity too. You can see I am also getting the quote lines and Account entities information here.



We are going to test the prompt now. I have created a test quote record in the dataverse to test the prompt now. 

  1. In the right pane in the prompt settings, select Input > Quote Summary Prompt

  2. In the Sample data field, enter sample prompt

  3. In the right pane in the prompt settings, select Data used (preview) > Quote, as shown in the following screenshot.

  4. In the right pane in the prompt settings, select Output > Text.


Now all set to go and test the prompt, click on test prompt . we can see the summary of the quote in the prompt response section

Now lets go and test this prompt in a cloud flow.

Start with manually trigger a flow and select AI builder>Create test with GPT using a prompt

We can see the prompt created earlier and enter the information. 
Add another compose action to check the output

Compose output 

NOTE: As you can observe in the above screenshot, the Flow checker shows the warning that appeared and in detail provides a way to fix it, we can skip fixing it but it is always recommended to incorporate the Approval after the “Create text with GPT using a prompt”.


Thank you for visiting the blog 🙏



Comments

Post a Comment

Popular posts from this blog

How to update 100000(hundred thousand) records with power automate flow by using chunk expression, Split the list and pass onto child flows.

-
Image
The Dataverse connecter returns up to 5000 rows by default, We can increase this up to 100000. To get more than 5,000 rows, turn on the Pagination and set the threshold up to 100,000 in Settings If you have more than 100,000 rows to process? you can get more than 100,000 rows from Dataverse, use the skip token to send another request until the skip token returns empty When we have thousand of records to update instead of doing list down the records and apply on each, I have approached this slightly different here.  We will build one parent flow and 4 child flows. So that we can divide the whole list into 4 identical list and pass on to child flows. This approach significantly reduces the time to update the total records.  Lets jump in and build a flows! Based on your requirement you can choose the type of flow you want, for example purpose I am using Schedule flow.  Since I am using the current date to filter down the records, I am initialize and format the date. use the ...
Read more

How to keep flows remain on if the flows don't run for 90 days

-
Image
This is a common requirement in power plat form world, flows turn off after 90 days if they don't run. We can address this issue by writing a scheduled flow and run every day.  I will explain how to write this flow.  Start with by selecting a schedule flow. Select Power automate management connecter- List flows as admin(V2) Once you select it will ask you to create a connection then Select your target Environment Next we need to filter the flows which we want to turn it on, Here I am using a unique display name which you have common for all the flows you want to check. In my case I am using "-Production-" to filter down the flows.  Select the value from the list flows Select edit in advance mode and enter the below expression @and(equals(item()?['properties/state'], 'Stopped'),contains(item()?['properties/displayname'],'-Production-')) In the next action, select apply on each , Select the output value from the filter array, Select the same ...
Read more

D365 CRM-Change Entity translations(Sub area) in Modern driven app.

-
Image
  Have you ever tried activating language packs in D365 CRM and tried to change the Entity names in Modern driven app? Normally When we activate the language pack for your preferred language Microsoft default translates all out of box entity names. If you want to change the translations for existing ones or create new for custom entities we need to update the translations.  You can enable the language pack under Administration>Languages Open the Modern driven app editor in classic version, select the Entity name you want to change the translations Select the language you want to update the translation and delete the existing translation and enter the new one.  Now the new translation name is updated.  Note: We cant update custom translations for recent and pinned fields in the left navigation at the moment. This is limitation from the Microsoft side at the moment.  Thank you for visiting this blog 🙏
Read more