However, it is important to remember that with the many benefits that AI brings, there are also potential risks and challenges that must be taken into consideration, particularly potential data breaches and privacy violations. Some companies shared their concerns about employees posting sensitive information to chat-based AI services and issued guidelines on how to use these services.<\/p>\n
As a Customer Success Manager, I\u2019ve seen several enterprise customers exploring solutions to give their employees access to the conversational AI in a more controlled and safer environment for the company. This is how I designed this solution, called MyAssist, based on Microsoft services: Teams, Power Virtual Agent and Azure OpenAI. MyAssist is deployed to employees as an application for Microsoft Teams.<\/p>\n
If you plan to deliver a bot service based on generative AI, I encourage the reader to review this page on responsible AI<\/a> to understand how this technology works, what you can expect from this type of services, what are the limits and educate users on how to prompt practices and practice for responsible use.<\/p>\n <\/p>\n Microsoft Teams<\/strong> offers a variety of security and compliance services to protect your data and help you meet regulatory requirements. These services include data encryption, role-based access control, multi-factor authentication, and compliance with industry standards and regulations. All chat messages (questions and answers) between a user and a bot application queried from Teams are handled as any other chat messages between users in Teams. Companies that implemented Security & Compliance service like Microsoft Purview via eDiscovery<\/a>, sensitive information types<\/a>, retention policies<\/a>, or other services like Teams Export APIs<\/a> or Safe Links for Teams<\/a>, and many others, can take advantage of these solutions for chat messages with bots in Teams.<\/p>\n Power Virtual Agents<\/strong> lets you create powerful AI-powered chatbots for a range of requests\u2014from providing simple answers to common questions to resolve issues requiring complex conversations. It is by far the fastest path to delivering a chat bot with built-in integration and publishing in Teams, including enabling authentication and user SSO in a click of a button. Power Virtual Agent has a large compliance offer<\/a> that includes HIPAA coverage, FedRamp, SOC, MTCS, CSA and more. PVA instances are deployed in your own Power Platform environment<\/a> to satisfy your security, user access, governance, data-loss prevention rules, authorized connectors, data residency requirements and more. Last, you can download and view up to seven days of conversation transcript sessions in PVA from the past 30 days (by default) and configure the transcript retention<\/a> to you needs.<\/p>\n Azure OpenAI Service<\/strong> gives customers advanced language AI with OpenAI GPT-4, GPT-3, Codex, and DALL-E models with the security and enterprise promise of Azure. Azure OpenAI co-develops the APIs with OpenAI, ensuring compatibility and a smooth transition from one to the other. With Azure OpenAI, customers get the security capabilities of Microsoft Azure while running the same models as OpenAI. Azure OpenAI offers private networking, regional availability, and responsible AI content filtering. With Azure OpenAI Service, customer can decide the location of the service, based on Azure regions and service availability<\/a>, and review Microsoft data, privacy, and security statements<\/a> on how data provided by you to the Azure OpenAI Service is processed, used, and stored.<\/p>\n By combining Microsoft Teams, Power Virtual Agent and Azure OpenAI Service, enterprises can provide a modern conversational AI experience, or any other generative AI model available on the Microsoft platform, that employees are so enthusiasts to use in a more controlled and secure way.<\/p>\n <\/p>\n All the services used in this article are still in (limited) public preview. It\u2019s important to remind the readers that features in preview have limitations:<\/p>\n For more information, go to our\u00a0preview terms<\/a>.<\/p>\n How to access these services:<\/p>\n <\/p>\n In this post, I document option #2<\/strong> that combines PVA and Azure OpenAI Service. Option #2 provides a good trade-off between the simplicity to deploy a bot in Teams via PVA, as opposed to a full development using the Azure bot service, and better control over the AI generative models via Azure OpenAI Service, as opposed to the AI Builder connector for ChatGPT on Azure OpenAI Service.<\/p>\n <\/p>\n <\/p>\n <\/p>\n <\/p>\n <\/p>\n <\/p>\n <\/p>\n Please check that you meet all the technical requirements and have an Azure subscription enabled to use the model gpt-35-turbo or gpt-4.<\/p>\n <\/p>\n <\/p>\n <\/p>\n <\/p>\n <\/p>\n <\/p>\n <\/p>\n Here is the JSON definition from the Shakespeare example on Azure OpenAI studio:<\/p>\n <\/p>\n {<\/p>\n “messages”: [<\/p>\n {<\/p>\n “role”: “system”,<\/p>\n “content”: “You are a Shakespearean writing assistant who speaks in a Shakespearean style. You help people come up with creative ideas and content like stories, poems, and songs that use Shakespearean style of writing style, including words like \\”thou\\” and \\”hath\u201d.\\nHere are some examples of Shakespeare’s style:\\n – Romeo, Romeo! Wherefore art thou Romeo?\\n – Love looks not with the eyes, but with the mind; and therefore, is winged Cupid painted blind.\\n – Shall I compare thee to a summer\u2019s day? Thou art more lovely and more temperate.”<\/p>\n <\/p>\n <\/p>\n <\/p>\n <\/p>\n Schema definition for the Parse JSON action:<\/p>\n <\/p>\n <\/p>\n <\/p>\n The result should look like this:<\/p>\n <\/p>\n JSON to be added in the \u201cEdit formula\u201d section of Power Virtual Agent in the Adaptive Card editor.<\/p>\n At this stage, you can test your bot in Power Virtual Agent and get its responses in an Adaptive card format. Once you\u2019re happy with the tests, from Power Virtual Agent:<\/p>\n <\/p>\n Your bot MyAssist running in Power Virtual Agent<\/p>\n <\/p>\n Your bot MyAssist running in Microsoft Teams<\/p>\nHow does this solution help mitigate the company\u2019s concerns?<\/h2>\n
Technical requirements<\/h2>\n
\n
\n
Design patterns<\/h2>\n
\n
![\"ChatGPT](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-1-1024x647.png\")
<\/a><\/p>\nLet\u2019s make it!<\/h2>\n
Step 1 \u2013 Create your instance of Power Virtual Agent<\/h3>\n
\n
![\"How](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-2.png\")
<\/a><\/p>\n\n
![\"Once](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-3.png\")
<\/a><\/p>\n\n
![\"Deactivate](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-4.png\")
<\/a><\/p>\n\n
![\"Once](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-5-1024x599.png\")
<\/a><\/p>\n\n
![\"Under](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-6.png\")
<\/a><\/p>\n\n
![\"Under](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-7.png\")
<\/a><\/p>\n\n
![\"You\u2019ll](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-8.png\")
<\/a><\/p>\n\n
![\"Test](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-9.png\")
<\/a><\/p>\nStep 2 \u2013 Deploy & configure Azure OpenAI Service<\/h3>\n
\n
![\"After](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-10.png\")
<\/a><\/p>\n\n
![\"Go](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-11.png\")
<\/a><\/p>\n\n
![\"Go](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-12-1024x669.png\")
<\/a><\/p>\n\n
![\"8.](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-13.png\")
<\/a><\/p>\n![\"8.](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-14.png\")
<\/a><\/p>\nStep 3 \u2013 Connect Power Virtual Agent to Azure OpenAI using Power Automate<\/h3>\n
\n
![\"Connect](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-15.png\")
<\/a><\/p>\n\n
![\"Delete](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-16.png\")
<\/a><\/p>\n\n
![\"From](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-17.png\")
<\/a><\/p>\n\n
![\"Add](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-18-660x1024.png\")
<\/a><\/p>\n\n
![\"6.](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-19.png\")
<\/a><\/p>\n{\r\n \"messages\": [\r\n {\r\n \"role\": \"system\",\r\n \"content\": \"You are a Shakespearean writing assistant who speaks in a Shakespearean style. You help people come up with creative ideas and content like stories, poems, and songs that use Shakespearean style of writing style, including words like \\\"thou\\\" and \\\"hath\u201d.\\nHere are some examples of Shakespeare's style:\\n - Romeo, Romeo! Wherefore art thou Romeo?\\n - Love looks not with the eyes, but with the mind; and therefore, is winged Cupid painted blind.\\n - Shall I compare thee to a summer\u2019s day? Thou art more lovely and more temperate.\"\r\n },\r\n {\r\n \"role\": \"user\",\r\n \"content\": \"Please write a short text turning down an invitation to dinner.\"\r\n },\r\n {\r\n \"role\": \"assistant\",\r\n \"content\": \"Dearest,\\nRegretfully, I must decline thy invitation.\\nPrior engagements call me hence. Apologies.\"\r\n },\r\n {\r\n \"role\": \"user\",\r\n \"content\": \"@{triggerBody()['text']}\"\r\n }\r\n ],\r\n \"temperature\": 1,\r\n \"top_p\": 0.95,\r\n \"frequency_penalty\": 0,\r\n \"presence_penalty\": 0,\r\n \"max_tokens\": 400,\r\n \"stop\": null\r\n}<\/code><\/pre>\n\n
\n
\n
\n
\n
![\"create](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-20.png\")
<\/a><\/p>\n\n
\n
\n
![\"initialize](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-21.png\")
<\/a><\/p>\n\n
\u201cSorry, I couldn't generate an answer from your prompt.<\/code>\nCan you rephrase your request?\u201d<\/code><\/li>\n<\/ol>\n![\"we](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-22.png\")
<\/a><\/p>\n\n
\n
body('Parse_JSON')['choices'][0]['message']['content']<\/code>Note: only the first answer generated by Azure OpenAI is extracted. Multiple responses are not supported in this pattern.<\/li>\n<\/ol>\n<\/li>\n<\/ol>\n![\"If](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-23.png\")
<\/a><\/p>\n{\r\n \"type\": \"object\",\r\n \"properties\": {\r\n \"choices\": {\r\n \"type\": \"array\",\r\n \"items\": {\r\n \"type\": \"object\",\r\n \"properties\": {\r\n \"finish_reason\": {},\r\n \"message\": {\r\n \"type\": \"object\",\r\n \"properties\": {\r\n \"content\": {\r\n \"type\": \"string\"\r\n }\r\n }\r\n }\r\n },\r\n \"required\": [\r\n \"finish_reason\",\r\n \"message\"\r\n ]\r\n }\r\n }\r\n }\r\n}<\/code><\/pre>\n\n
![\"Configure](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-24.png\")
<\/a><\/p>\n\n
![\"12.](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-25.png\")
<\/a><\/p>\n\n
\n
![\"Back](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-26.png\")
<\/a><\/p>\n\n
![\"Add](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-27.png\")
<\/a><\/p>\n{\r\n 'type': \"AdaptiveCard\",\r\n '$schema': \"http:\/\/adaptivecards.io\/schemas\/adaptive-card.json\",\r\n 'version': \"1.3\",\r\n 'body': [\r\n {\r\n 'type': \"TextBlock\",\r\n 'text': \"Answer generated by Azure OpenAI using ChatGPT model \",\r\n 'wrap': true,\r\n 'size': \"Medium\",\r\n 'weight': \"Bolder\"\r\n },\r\n {\r\n 'type': \"Container\",\r\n 'items': [\r\n {\r\n 'type': \"TextBlock\",\r\n 'text': Text(Topic.response),\r\n 'wrap': true\r\n }\r\n ],\r\n 'style': \"emphasis\"\r\n },\r\n {\r\n 'type': \"TextBlock\",\r\n 'text': \"NOTE: This service is in preview version and model does not have context of previous messages.\",\r\n 'wrap': true,\r\n 'color': \"Accent\",\r\n 'size': \"Small\"\r\n }\r\n ]\r\n}<\/code><\/pre>\n\n
Step 4 \u2013 Test and deploy to users<\/h3>\n
\n
![\"test](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-28.png\")
<\/a><\/p>\n![\"Your](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-29.png\")
<\/a><\/p>\n
![\"Your](\"https:\/\/devblogs.microsoft.com\/microsoft365dev\/wp-content\/uploads\/sites\/73\/2023\/04\/ChatGPT-blog_image-30.png\")
<\/a><\/p>\n