Git Repo: https://github.com/dtsoden/ChatAssistant
Chat Assistant - Human Centricity
Features
- Finally, a secure way to implement ChatGPT in Enterprise.
- Easy to use, end user Interface.
- Rating system for each AI Generated Response.
- Feedback System.
- Users can create multiple prompt personas to suit and how they work.
- Self-Monitoring, Search logs of past interactions to encourage reuse.
- Built-in & Secure Administrative Backend
- Live Auditing Dashboard.
- Admin Feedback Log
- In App Moderations, “BLACK List Words” prevent submission, “GREY List Words” are logged.
- Notifications of Feedback submissions and Triggered “Grey List Words”
Requirements
- Power Platform Premium Licensing
- Dataverse
- Custom Connectors
- Azure Subscription
Azure and Power Platform Technologies Used
- Azure Cognitive Services, OpenAI
- Power Platform Custom Connector – connects Azure to the Power App
- Power Platform Canvas Application – the application
- Power Platform Power Automate – the notification system
Azure OpenAI, Access, Limits & Requests
- 1st you may need to request access to Azure OpenAI Services. If you need to access this link and fill out the form. https://aka.ms/oaiapply
- This service has a hard set of limits that may become restrictive as this solution scales. If you reach limits the application will error out often or not work at all. You can request increases here https://aka.ms/oai/quotaincrease to avoid this.
--------------------------------------------------------------------
This is your opening screen to the application. You are given a friendly reminder that clicking the logo will start a new Chat session. This banner is temporarily displayed and disappears after 5 seconds.
--------------------------------------------------------------------
In this screen you can see a full Chat turn where in the user has asked a question and the AI has responded. Several things to notate on this screen. Users can copy the AI response directly AND from within the chat history in the lower left. Users can also copy their prompts as well. Users can vote on the AI Response and provide feedback. More on this later. Take notes that above the Chat box it says “Default Profile”, this is the current chosen profile that is active. More on this later
--------------------------------------------------------------------
In this screen the user has pressed the Orange FEEDBACK button and is presented with the dialog to enter in their feedback. Doing so will trigger a Power Automate email notification to the configured recipients from within the Cloud Flow.
--------------------------------------------------------------------
Here is one of the most powerful features Prompt Engineering Persona Profiles. Users often wear different hats to perform the various roles expected of them, and this feature fits perfectly. Users can have personas for general use, Executive writings, and Sales emails. There is no limit to how many Persona Profiles a user can have.
--------------------------------------------------------------------
Here Users can access, see and copy out past prompts and the responses from the AI. They can also see how many tokens they have used up to date.
--------------------------------------------------------------------
The menu system is dynamically configured. Administrators of the system will see the full menu, where as regular users will not see menu items, specifically designated for admins.
--------------------------------------------------------------------
On this administration screen, lies a powerful dashboard that gives you full governance and compliance auditability. See all past chats, click a row to see the entries tetails with full copy capabilities. See other stats at a glance such as token consumption across all users at the bottom of the screen or see your top 10 users. Finally you get a snapshot over the ratings of the AI Responses.
--------------------------------------------------------------------
Here are the details of a spcific row from witin the Dashboard. You can copy the Prompt and the AI Response.
--------------------------------------------------------------------
In this administrative screen admins can review the full feedback log in context to the Prompt and AI Response. Remember you also get an email notification for each time feedback is submitted.
--------------------------------------------------------------------
In this screen you can configure words you wish to moderate by flaging them as BLACK = Forbidden, or GREY allowed but flagged and another notification is sent to those whom are configured from inside the Cloud Flow.
--------------------------------------------------------------------
Download the Attachments below for screenshots and more
