ChatGPT – Ben Prins

Power Platform: Use AI to evaluate every incoming email

November 11, 2024 Ben Prins Power Platform

Leveraging AI for email management not only saves time but also reduces the risk of human error. By automating the evaluation process, businesses can ensure that important emails, such as orders, are promptly and accurately identified, allowing for quicker response times and improved customer satisfaction. Embracing AI in email workflows is a smart move towards greater productivity and operational excellence.

In this article, I explain how to create an AI prompt model using Microsoft GPT to streamline email processing. By integrating Power Automate, the flow sends the content of each new email to the AI model, which then determines if the email is an order. If identified as an order, the AI responds with a JSON output indicating a positive result. This approach enhances the efficiency and accuracy of order processing through intelligent automation.

Create an AI model

For this blog I create a simple AI prompt to evaluate if a incoming email is a request for an order. The actual prompt is more complicated but confidential.

Open the AI hub aka AI Builder and select the Create text with GPT using a prompt.

Give the prompt a name.
Create three input parameters.
- Email
- EmailSubject
- Attachments

Select by Output JSON (preview) and click on Edit.

Add the following JSON code.

{
      "Order": "true"
}

Set the Model to either GPT 3.5 (cheaper but less accurate) or GPT 4 (more expensive and more accurate).
Set the Temperature to 0.

Set the Prompt as follows or make your own prompt.

Save the prompt.

Create a Power Automate Flow

The Power Automate flow will start for every incoming email and will send to the AI model, the email body, subject and names of the attachments. The model will return its determination in JSON form.

Create a new Power Automate flow with an outlook/email trigger, When a new email arrives in a shared mailbox.
Add an Initialize variable action to create a variable called All attachment names.

Add an Append to string variable and select by the name the All attachment names variable.
Set for the value the attachments of the email, with the following code. This will automatically add a Apply to each action.

items('Apply_to_each_-_Attachment')?['name'] -

Add the Create text with GPT using a prompt action.
Select your created AI Prompt.
Set by attachments the variable All attachment names.
Set by Email the Body of the email.
Set by EmailSubject the Subject of the email.

This will start the AI model and it will return its results to the flow.
Form here on you can add in the requered steps for your specific process and test the process.

Power Automate: Set email category

October 28, 2024 Ben Prins Power Platform

In a recent AI project, I developed a solution to evaluate every incoming email in a shared mailbox using an AI model (GPT with prompt). To inform users of the shared mailbox that an email has been processed, the email is flagged with a custom Outlook category. This process leverages a special HTTP action for Outlook, simplifying the implementation. Notably, there is no need to set up any special permissions if the flow owner already has access to the shared mailbox. This example does not cover the specifics of communicating with the AI through Power Automate.

Creating custom category in Outlook

Open Microsoft Outlook.
Click on Category in the Home Ribbon.

Click on All Categories.
Click on New.
Set a name and select a color.

Click on OK followed by Clicking on OK.
You have now created the custom Outlook category.

Setting the category with Power Automate

Create a new Power Automate flow with an outlook/email trigger. For example: When a new email arrives in a shared mailbox.
Add the Send an HTTP request action.
Set the URI to be following code.

https://graph.microsoft.com/v1.0/me/mailFolders/Inbox/messages/triggerOutputs()?['body/id']

Set the method to PATCH
Use the following JSON code for the body

{
"categories": ["AI Finished"]
}

Set the Content-Type to application/json.

Save the flow.

Power Platform and Chat GPT

March 2, 2023 Ben Prins Power Platform

In today’s digital age, businesses are constantly looking for ways to streamline their processes and improve their customer experience. One way to do this is by leveraging the power of chatbots, which can quickly and efficiently answer customer inquiries. In this blog post, we will explore how to create a canvas app that uses Power Automate flow to ask ChatGPT API questions and display the response in the canvas app. By the end of this tutorial, you will have the tools and knowledge to build your own chatbot app that can answer your customers’ questions in real-time, enhancing their overall experience and increasing your operational efficiency. So, let’s get started!

Create a Chat GPT Api secret

Open the ChatGPT API site.
Login or create an account.
Click on Personal, followed by View API keys.

Click on Create new secret key and save the key in a password vault.

Creating the Power Automate Flow

Create a new Power Automate flow with the name Canvas app – Chat GPT.
Add as the trigger a PowerApps V2.
Add a text input with the name Question.

Add a HTTP action with the name Post to Chat GPT.
Set the Method to POST.
Set the URI to https://api.openai.com/v1/completions.
Set the header to Content-Type with value application/json.
Set a second header to Authorization with the value Bearer [API Secret].
Set the body to the following json code.

{
  "model": "text-davinci-003",
  "prompt": "triggerBody()['text']",
  "temperature": 0,
  "max_tokens": 4000
}

Add a Parse JSON action with the name Chat GPT Response.
Set the Content to Body (response of the HTTP call).
Set the following schema (update the schema is the response is different).

{
    "type": "object",
    "properties": {
        "id": {
            "type": "string"
        },
        "object": {
            "type": "string"
        },
        "created": {
            "type": "integer"
        },
        "model": {
            "type": "string"
        },
        "choices": {
            "type": "array",
            "items": {
                "type": "object",
                "properties": {
                    "text": {
                        "type": "string"
                    },
                    "index": {
                        "type": "integer"
                    },
                    "logprobs": {},
                    "finish_reason": {
                        "type": "string"
                    }
                },
                "required": [
                    "text",
                    "index",
                    "logprobs",
                    "finish_reason"
                ]
            }
        },
        "usage": {
            "type": "object",
            "properties": {
                "prompt_tokens": {
                    "type": "integer"
                },
                "completion_tokens": {
                    "type": "integer"
                },
                "total_tokens": {
                    "type": "integer"
                }
            }
        }
    }
}

Add the Respond to a PowerApp or Flow action.
Add a text output called ChatGPTRepsonse and add the response from Chat GPT with the following code.

first(body('HTTP_-_Post_to_Chat_GPT')?['choices'])?['text']

The overall Power Automate flow will look like this.

Creating the Canvas app

Open the Power Apps Studio and create a new canvas app.
Rename Screen1 to Home.
Add the Canvas app – Chat GPT Power Automate flow to the canvas app.

Add a Rectange Shape to the top of the canvas app with the name RectTitle.
Add a label over the RectTitle with the name lblChatGPT.
Set the Text to “Send your question to the all powerful Chat GPT AI bot”.
Add a label with the name lblGPTRepsonse.
Place the lblGPTRepsonse on the right side of the screen.
Add a text input with the name txtQuestion.
Place the txtQuestion on the left side of the screen.
Set the txtQuestion Default to “What is your question?”.
If you like add an Image with the name imgRobot and add an image of a robot in the Image property.
Place the imgRobot left and next to the lblGPTRepsonse.
Add a button with the name btnSendQuestion.
Set the following code on the Onselect of the btnSendQuestion.
- This will save the response in the variable repsonsegpt.
- Start the flow with the text provided in the textQuestion text input box.

Set(responsegpt, 'Canvasapp-ChatGPT'.Run(txtQuestion.Text).chatgptrespondse)