Introducing Custom Engine agents

Key Takeaways

imagine building an agent that works on Microsoft 365 uses the AI model of your choice runs with customer or castration leverages retrieval augmented generation with Azure Integrations I mean really really custom one that my developer friends are now called custom engine agents hi my name is aab Bash I'm Cloud developer Advocate at Microsoft in this video we're going to break down custom engine agents if you want to learn anything and everything about custom engine agents keep on watching let's Di in I want to show you this awesome Human Resources expert career Genie career Genie is a b framework based agent that works on teams it's powered by team's AI library and Azure open AI it's friendly and helps with HR queries as you can see but let's have a closer look to its user experience first thing we recognize is the AI generated label and we also have the sensitivity label right next to it we also have feedback loop with thumbs up and thumbs down buttons where we can share more feedback about what we like and dislike about this answer this all really looks and feels like a co-pilot experience we can also ask career Genie complex questions like suggest me. net developers who can speak Spanish as career Genie is connected to Azure AI search and can apply Vector search on top of a diverse set of rumes we can get the best candidate suggestions for our specific query career gening can do much more than just retrieval augmented generation with Azure AI search to start with it can authenticate users with single sign on and it can handle much more complex tasks for you for example when we get some candidate suggestions we can actually create list of candidates using career Genie and career Genie understands the task thanks to the natural language processing and takes an action accordingly recognize that it can also handle complex queries like add Anthony in the same list with Isaac and easily get the task done it can summarize our list and also can create multiple lists let's create another list for python developers and add Sarah in it can also keep track of changes in the list for example we can remove Anthony from the net list and add Anthony in the new Talent list we can keep making many other changes and career joury will be able to keep track of that it will be able to create much more lists than just three it can add more candidates in the list and be able to make all these adjustments in real time and once we are settled with our list of candidates we can ask send my list to HR and it will be able to send my list to HR for scheduling interviews if I quickly go ahead and check my inbox I should be able to see lists of candidates sent to HR for scheduling interviews career Genie definitely looks cool but I'm pretty sure you already have some questions how can we make career Genie work so smoothly on teams how does it handle diverse data set with Rag and have this awesome characteristic Behavior also have co-pilot like user experience let me show you how our custom engine agent career journey is powered by teams Library let's go through the core components in our source code let's start with app. TS we start with the AI components here first we have the model we want to use open AI model helps us call the Azure open AI then prompt manager manages prompt creation using the files under the prompts folder then we have the action planner that uses large language model to generate plans and calls Azure open Ai and finally application utilizes all of these AI components like model planner and so on and replaces the activity Handler class in a typical bot framework bot let's have a look at the prompts folder under the prompts folder we have chat folder then we have config.js that has all the settings about our prompt and then we have skap prompt. txd that keeps the meta prompt and defines the behavior of our bot in this case it's career Genie that helps Human Resources team finding right candidates it's friendly professional likes using emojis and so on so far these are the core components of a custom engine agent let's take it one step further and see how we can include Azure AI search as a data source if you go to config.js you will recognize that we have data sources here and this is the place where we Define Azure AI search as a data source and if you look at the details here you will see asure AI search endpoint index name key cury type as vector and embedding model all right the next step is customizing the user experience using powered by AI kit in teams AI Library I'll go back to app. TS here the first feature I want to show you is feedback R you can add thumbs up and down buttons in each response of the AI for users to share their feedback and then the citations you can actually customize the citations using predicted say command and adaptive card another great feature available in powered by AI kit is generated by AI label where we can add in each AI response to highlight that this is generated by Ai and finally we have the sensitivity label where we can Define the confidentiality of the data are AI is sharing let's continue with another important capability we can add in our custom engine agents which is authentication you can use the authentication component available in teams library and enable single sign on finally let's talk about actions which is one of the most important components that adds the complexity in our conversation with AI to handle complex tasks like creating lists adding candidates making changes adjustments to handle all these actions we will need to create another prompt that means our default prompt will be choosing in between two different prompts the monologue folder under the prompts folder handles all the actions related tasks in the actions. Json we Define all the actions here such as create list delete list add candidates remote candidates s lists that is for HR and then we will need to register actions in our app our Ai and planner will decide which action to call and once we hit the action we will be calling the related function available in the AC actions. TS all right how do you build one of these from scratch well we are going to use teams toet in vs code select team toit icon then create a new app you should select custom engine agent and then basic AI chatbot select the language of your choice aure open AI enter your keys and point deployment name select the default folder give your application a name then we are ready to go to have the basic functionalities of career Genie the first thing you should do is to go to skape from txc update here with career Genie's behavior and then hit F5 if you're trying this for the first time you can use teams app test tool which will be a lot easier to test and see the results and because we don't currently have Azure AI search integration and all these complex capabilities we should start by asking simple question like can you help me write a job post with senior developer role make sure C is a part of it and career gen is already helping us right away are you already interested in building one I have a amazing news for you we are introducing a new virtual Workshop called co-pilot developer camp we have series of labs where you can follow to build your own custom engine agent AKA career Genie with its capabilities including rag powered by AI authentication actions and much more check out co-pilot developer camp today to start building your own custom engine agent thanks for watching And subscribe to our channel for more co-pilot development content See you in the next video bye