Hi, everyone. Welcome to this Conf42 Prompt Engineering Conference. And even more welcome to my session topic on exploring co pilot capabilities. Now, for today, we have, about 45 minutes presentation time, where I'm going to try and show you quite some demos to keep it a little bit entertaining. Now, in short, although the main topic of the conference is prompt engineering, I actually wanted to take it one step further and talking about how to actually use prompt engineering using Microsoft Copilot and not only in like web or work like M365, but actually giving you a walkthrough of what it different co pilots, are available today within Microsoft. Now I'm going to start with a quick overview, what generative AI is about. Next to that, I'm going to zoom in a little bit on, obviously, the prompt engineering, and then from there, walking you through a couple of scenarios, where you already know, or should know, that the ones I'm going to walk you through are not the only ones available in, the Microsoft world today. Now, before we're going to jump into the technical details, I'll share a bit about myself. So Peter De Tender, originally from Belgium, but moved to Microsoft Redmond close to three years ago. I've been working for Microsoft a bit more than five years as a Microsoft Technical Trainer, which means I'm basically providing full day sessions, sometimes full week sessions, workshops on the Azure platform where my main technology focus has always been Azure architecting DevOps and in meantime, app modernization using developing on Azure as well as co pilot and AI. If you should have any questions during the conference, feel free to reach out. If you're watching this later on as a recording, nothing should stop you from reaching out. I got my LinkedIn details up here. You can reach me on Twitter. You can send me an email. So now let's jump in and talk a little bit about Generative AI. So these are just a couple of headlines from, a bit more than a year ago, actually. Where everything was around, revolutionizing AI using the chat GPT generative AI capabilities. And there's a lot of headlines here, but the main goal is that it's here. Yes, a bit more than a year ago, it was totally new. Nobody really talked about it. And I would say the capabilities are crazy amazing. And it's also a big part of what I want to share today. Now, giving you one easy example, this is where, for example, I started using. A prompt to generate my bio picture, like when I'm starting training, like every morning basically, or sometimes like Monday morning in a multi day class, then I need to talk a little bit about my background, like building up credibility for the training. And that's where now I start playing a little bit with prompt engineering. To generate my bio picture. Now, the thing is, how do you get from your image into a prompt engineer component? So I'll walk you through the basics where it should not be as harder as what I'm going to try and show you. If you look a little bit at this picture, you're going to see that there are a few different elements. And I would say for me, just as an example here. What I would highlight is, for example, Seattle. Now you know my roots, it's Belgium. So you can see the Belgium flag up here. If we look at Peter himself, you can detect a little bit of spiky hair. You can detect glasses. You can detect the hoodie which I like to wear. You can detect the Microsoft logo and some sort of a laptop. Just looking at that picture and recognizing the key components, that's all you need to build up. Your prompts, why? Because if I move all this into some few sample prompts actually used for generating those images, the first one was cartoon man, Pixar style. So you can use, some already existing, artistry, you could say. But now the rest of my prompt across all three is basically the same with a big smile wearing glasses, thin beard, dark brown, short hair, gray hoodie, Microsoft logo, Surface laptop in the background, show the Seattle skyline and the Belgian flag on top of it and just reusing like 99 percent of the same prompt, but then only making some minor changes. Flipping from Pixar style to Simpsons style to Marvel comic style. That's how you're going to end up with pretty cool and sometimes even remarkable, impressive, outcomes of that. Now that's the setting the scene on prompt engineering. Now in short, what is prompt engineering? It's where, again, about a year ago, The world was going crazy. So again, just a few headlines here from, reputable magazines and news, papers where I picked up like, it's actually a pretty cool thing. So a lot of them literally talk about like the first one here, rising significance in prompt engineering, or the other one, the number one thing CEOs are investing in right now is, guess what? Generative AI. And then the last one here, obviously the one that I like most is that thanks to generative AI. The AI job paying more than 300, 000 without having a computer engineering background. quite impressive, but is it all really that important and that easy? in short, what is prompt engineering? The first part in engineering is, prompt engineering is the prompt. a prompt can be described as, involving instructions, providing details, providing a context, Passing it on to a large language model and then achieving a desired task. The second part is engineering. When our engineering here refers to developing, optimizing prompts to efficiently again, use that large language model. Now, by the way, if you want to learn more prompt engineering, I'm going to give you another easy example. Now, I would say prompt engineering sounds a little bit wrong. Now, why? Because engineering sounds hard. It sounds difficult. And since not all of us are engineers, it almost seems I'm not even able to use that generative AI technology. So that's why within Microsoft, we actually like to use the term the art of the prompt, or as I like to describe it, prompt artistry. Now, what does it mean? You could grab like a tool of brushes, you could grab some colors, you could have a canvas, but that's no guarantee to actually come up with a nice painting. On the other side, prompt artistry should not be all that hard, since in the end, we have the natural language concept, allowing us to assist with, creating our prompts, but also using it. And in the end, you're basically switching. And that's the part that I like about it. You're switching from searching into finding. And the way to do that is basically asking a series of questions. So moving this into a business context that later on, I'll show you in one of the demos, you'll see that. A scenario could be something like this. We have a goal in mind, which means what do you want Copilot to do? Next, it had the context, like why do you need it? who's going to be involved? Apart from that, we have the source, where later on we'll talk about web and work, and eventually what are the expectations. So the example here could be, I need to review, come up with details about Microsoft Financial Report. The context, I need it for a meeting that I have next week with Microsoft executives, for example. The source could be web or work, which means I'm going to use the trained language model information, like public data that it's been trained on, Or, and that's specific to Copilot in M365 that I'll share about, a lot more later on, using your corporate, Microsoft 365 backend data like OneDrive, email, Teams, and SharePoint. And from there, my expectation is I don't need all the details, but I want you to give me like a summary in seven bullet points. It's almost identical to approaching Your colleague approaching the CFO, maybe in this case, oh, could you provide me like the Microsoft report? But instead of giving you all the details, you're gonna fine tune it, you're gonna summarize it, and then the actual prompt could be something like this, where everything is now nicely coming together in one larger prompt, where the prompt, again, is technically almost identical to just asking a question in the, I dunno, human world without using copilot to help you with that. Now enough talking about it, I'm going to switch to, my first demo here. So what I'm using right now is copilot. microsoft. com. So you open up your browser. You connect to copilot. microsoft. com and it's going to ask you like, Hey, how do you want to authenticate? Do you want to use work? So that's the difference web and work. Or do you want to use a personal account? Now I'm going to switch to my other window. And actually trying to show you, there we go, what it looks like in the personal space. Now, this is what we describe as unlicensed. So I'm authenticated with my Peter at Google account, and I could ask basically any prompt. There's a few sample prompts on screen, and I could do something like this. What, Oh, I already have it here. So I'm, my roots, I'm from Belgium, which means I like food. I like beer and preferably food that has some sort of beer in it, which means, beef stew. So to come up with some new ideas, I literally. Went for a prompt. I'll show you in a second. I got beef stew meat. Like I went to the store, got me some beef stew meat. And can you just come up with some ideas, share a few recipes. So without doing anything specific or spectacular, it's now coming up with a recipe. a couple of different recipes. It shows me the ingredients and it also provides me the actual instructions, how to, start creating this. Next to that, I could also do something like, can you create an image showing a delicious beef stew meal? And that's what I could maybe, for example. Show to my wife later tonight when she's back from work. Back to our corporate environment. So the entry point is the same. I'm using copilot. microsoft. com. We're now instead of using personal, I'm going to switch to my work environment. What this is going to do is shifting to what we now call biz chat. So what is biz chat? There's actually a short link. M365. cloud. Microsoft, but that's a little bit new, I would say because before we all had to go to copilot. microsoft. com. I'm in my trusted M365 environment, where now it means I can actually start finding information, using information, and then later on maybe editing my environment. Now, before that, I'm going to switch back to the presentation for a little bit. And talking about all the other cool stuff inside Copilot. Now, in short, what is Copilot? Copilot is like a collection, you could say, of different AI assistants. Powered by the generative AI to help you with what we call cognitive tasks. Copilots are often driven by natural language. The large language model that I already talked about. Copilots are also embodied by a chatbot. So that's what I showed you in one of my first demos. Humans are still in control. That's one of the amazing things I like about the term co pilot is that you're still the pilot. You're in charge. You need to validate the answers and you're going to decide eventually what part of it that you're going to use. Now, the main value I would also say is really depending on what kind of tasks are you performing and I'll walk you through a couple of demos on that. that later on. Now we describe it as the era of co pilots. Now what does it mean? Think of co pilot as that well trained, well aware, up to date, real time collaborator for basically anything. So it allows you to generate content. It allows you to literally put Become that assistant if you want one of the building blocks. By the way, I didn't mention in that overview. what is a good prompt is a persona and I'll show you that in a demo later on where you could now go Hey, co pilot, you're acting as an Azure technical trainer. And you can only answer Azure technical questions. If there's any question coming up about, how to prepare beef stew, you won't be able to, answer that prompt. Or, obviously, flipping it around, like your, I don't know, 15 year experienced, I don't know, high quality Michelin star alike restaurant chef. You've been managing the kitchen for 15 years and I want you to come up with a delicious meal, recipe and so on. But you do not know anything outside of the kitchen. Something like that. So that's another cool thing. So it's again back to our co pilots. So it's going to allow you to generate content. It's going to allow you to optimize your productivity. It's going to allow you to automate your cognitive tasks and then eventually, helping you completing your tasks in a more fun way. Now, although I called my topic copilot capabilities, it's actually not even possible anymore to describe even the basics of copilot or all different Microsoft copilots we have in meantime. I'm going to try and zoom in a little bit on my screen here, but if we look at like the capabilities, Peter is big over there, so I would say we have co pilot. We have capability integrations for developers for business applications. That would now be our dynamics environment, and then we have, Modern work, so everything around Microsoft 365, data and AI, that's the core of our AI solutions, still integrating with machine, machine learning using the Azure platform. In the browser, a little bit, whoops, what I already showed you. And the main message here is, there's a copilot for pretty much anything, for pretty much any user. And that's what this slide is actually trying to jump in every now and then. Maybe something wrong with the animations, but this is a slide that I, a slide, sorry, that I personally like a lot. Why I call it the rooms of the house. This could be your private environment, obviously your apartment, your home, where you could talk to your wife, your partner, your kids, and stimulating them to use Copilot for pretty much anything. Now, obviously this is a little bit more in the business context. And what it means is again that we do have copilots for pretty much any user. If you store corporate data inside SharePoint, Teams, Outlook, which I think everyone is pretty much doing in the Microsoft world, you're going to rely on copilot for Microsoft 365. If we look at the business part of it, like where you're going to store business data, it might be. Co pilot for Dynamics, where we're now covering solutions for HR sales teams, marketing teams. That's the idea. If you're a sysadmin, if you're a data engineer, you're going to look into managing co pilot integration using co pilot for Azure. You have co pilot for security, co pilot for Azure, where it's going to help you looking into your Azure subscriptions, your Azure environments, and providing you feedback like how are. My service doing, can you help me creating a new Azure template to do something? That's the idea. If you're a developer, you're going to focus on co pilot for GitHub or now, I think the official term is GitHub co pilot. I'll walk you through that a little bit later on, and then eventually co pilot for anyone else where you might go for power apps or co pilot studio, allowing basically any user. to create their own custom copilots outside of what's already available here on screen as a capability. So from here, I'm going to walk you through a few different examples. I'm going to start with Copilot 365, later on GitHub Copilot, and then eventually showing you apart from the capabilities, I'm going to walk you of Copilot in M365, Copilot in GitHub, how you can actually create your own Copilots, whether you're a developer or just a traditional user. And hopefully that's going to become clear. So the first piece is adopting Copilot. So imagine the story that I like to use in this presentation is, we started with generative AI. Maybe you looked into ChatGPT or maybe, You're still using it today, where now I'm going to try and convince you that there are some additional benefits in using M365. So my first wave here is adopting Copilot. We talked about the private space, the unlicensed space, but I would say for corporate use, preferably you're going to pay for that Copilot M365 license. I think roughly, but not all that important. Check with your Microsoft account team, I would say. But I think on average, it's around 30 per user per month. Now you might think okay, Peter, that's a quite a big investment for a larger organization. So what's the gain? First of all, you have access to the large language model. In web and work. I showed you a super basic prompt, give me some recipes, around beef stew in the private space, and I'm going to show you the same prompt that, or similar prompt that I'm going to use in the corporate space next to that. It's going to allow you to use generative content creation. You can use Copilot in M365 in all Office applications. In Word, in Excel, in PowerPoint, in OneNote, Copilot pages, one of the newer capabilities. You can use it to find information, you can use it to brainstorm an idea, and again, basically using it as that well trained colleague. And then in the end, optimizing productivity. So in short, how does it work a little bit from like a helicopter architect perspective? You have the large language models. Copilot in 365 is using the GPT 4 model. So one of the most up to date trained models. It's going to hook into Microsoft Graph, where Graph is the API integration that we're using to present you data from Office 365. One drive, one machine. SharePoint, emails, and Teams. Now, the other nice thing is that although Copilot runs as a service, it's embedded in almost each and every Office product in the meantime, but it's also respecting your security. Which means that it's going to run in your usual account mode. So your data is your data. We're not using it to train where some other generative AI solutions are actually using your prompts to train the models and they're using metadata from your content to again train the model. Within Copilot we're not doing that. So your data remains your data and it's also respecting security boundaries. It's already mentioned integrated in M365 applications, and you can also use it in the web what I showed you in my previous demo. So that let's shift to, the next demo here, where I'm going to go back to what we now describe as the Co Pilot M365 BizChat. And that's over here. So I landed in copilot. microsoft. com. Do you want to use personal? Do you want to use work? And we land here. A lot of different options you can do, or you can use from here. So it's going to try and help you a bit with, sample prompts. The sample prompts here are actually, coming and I'll show you that later on. From what we call co pilot lab. So if you do aka. ms forward slash co pilot lab, it's going to show you how, you can come up with what I call sample prompts. They've been created by the product teams vetted by, the first wave of users and feedback from customers as well. So for example, within the corporate space, you might need help with preparing for a meeting. You might be looking for a file type, or a document name that you remember from like a customer, but you don't really know where to find that document because it's somewhere in your SharePoint, it's somewhere. In your OneDrive, you know that it's there, but you don't really know exactly where to find it. And that's where your copilot is going to help you. So I'm going to go back to, another sample idea here. Let's say in preparation for my meeting, can you. I'm going to go ahead and list up the latest Microsoft financial report. Now, before I'm going to run this, let me talk a second about work and web. So thinking back about the previous slide, I talked about the large language model using public trained data. That would be the web. We're now, if we switch to work, that's where it's using your corporate M365. So depending on the task, remember you're in control. You need to think about where. Am I going to find the information that I need? So if I use the web tab, for example, and I'm going to run this prompt, what it's going to do is pulling up information. Again, this is not a hundred percent real time, internet search, but it's providing me feedback based on what the large language model has been trained on. Now it's going to provide me here, some input, all good. And it's going to show me the actual annual report. So that's another nice thing with Copilots that it's going to show you not only the prompt feedback, the response, but it actually also comes up with, the source where that information came from. So I could use. Like the web page here, and that's going to show me all the details and I can download it. And that's what I'll reuse later on. Now, the nice thing is here. Like the sources are public web, sorry, public internet pages. now, if I switch to work, what it actually means now is that I'm going to use the same prompt, but this time it's going to use internal corporate sources. SharePoint, OneDrive, Teams meetings, for example, and it's typically also providing more insights, more details because it's closer to the employees. Now, the baseline scenario is still the same. So it's going to give me the details, but it's also pointing to files that I have access to. I can shift to Copilot in Word, for example, where now this idea is I downloaded that Microsoft Financial Report and I can have Another integration with Copilot. There's actually two different ways to use this. So one example is what we call Copilot drafting. So whenever you're in a Word document, in PowerPoint, in Excel, in OneNote, you're going to see this little Copilot icon showing up over here. What it allows me to do is now using Copilot generative AI capabilities to help me with my document creation, my document research, my Business idea, compilation and whatnot. So I could do something like writing a prompt back to the basics, where now I could do transform this into a table. The use case here is that maybe I don't know how to do this in Word, but I like to find that information, reuse it and move it into a table. That's what, again, what we call drafting. Why? Because. It's still a draft. I need to confirm keeping it or I could go back to the prompt and actually updating my prompt. I'm going to keep this change for now. That's all fine. Now, another use case is using Copilot to help me with my document, but this time I'm not really changing anything in the document, but I'm going to take some other actions. Again, we do have our Copilot lab based, sample prompts to get started. And if you want more. You can actually use this view prompt, and it's going to show you the copilot lab vetted prompts from here based on the task that you want to perform. The prompt scenarios are different in Word, in PowerPoint, in Excel. That's the idea. So there is a prompt here. Summarize this document in X amount of points. Now, again, showing you that it's understanding your natural language. You could do something like, can you provide. Need, key takeaways from this document. I don't need to add from this document because I'm already inside the document. Now, I like this one. Key takeaways because it might have a different connotation for basically, any different user. If I'm in a financial report and I would ask my CFO, can you provide me the key takeaways? The answer will be totally different than if you would ask, one of your HR people, or if you would ask, like in my case, a technical trainer. Can you talk about Microsoft Financial Report? The answer will probably be slightly different. But even then, the core concept remains the same. It's going to provide me the prompt input, it's going to provide the response, and it's also providing me the source. This time, no source on the Internet, but it's a source within the Word document, and then jumping across different paragraphs. Another cool example, is inside PowerPoint. One of the things that we all have to do, like creating PowerPoint decks for our presentations, right? Where I'm going to start from a new presentation, where again, you can see up here, we do have Copilot and here on the side again as well. The use case is again, slightly different depending on again, the task you want to perform. So I'm going to start the upper one here, and it's going to offer me what we now call the CoPilot PowerPoint Narrative Builder. Imagine my scenario here. If you paid attention to the pitch at the start of the session, it's like making users more productive. So the flow would typically be, doing research, going to the internet, finding information, go to SharePoint, OneDrive, anywhere on the planet. To find your information, you're typically going to move that into a Word document and building up your flow, or your story of your presentation. And then eventually you're going to move all that into PowerPoint slides. All that can now be optimized with one easy prompt. So I'm going to reuse the same prompt, create a presentation about, and so on. I'm using the exact same prompt, copy pasting. And again, it's going to use generative AI. To build up the layout of my presentation based on the content, the actual headlines, the chapters, if you want, within my financial report, based on the same generative AI prompt that I used before. I still have. Co pilot integration, I could add something at a topic about Azure AI. I could move my outline components around, so I could move it up and down, and then eventually asking to generate the actual content. What it's going to do here is starting with the base outline, it's going to create the different slides. Next to that, it's going to add It's going to add all the different chapters and the last it's going to add content where it's not just some minimal text, but it's actually providing full details, adding images, adding bullet points. And on top of that, adding speaker notes as well. And all this in, again, just a couple of minutes. I did not edit this piece of the presentation for the recording, by the way. If I open up one of these slides. You can see that everything is there. The information is coming from the prompt response and it's also providing the speaker notes. The other nice thing is from here, if I go to co pilot on the site, remember I briefly talked about graph. So what I can do is add information from forward slash and that's going to allow me to browse for files. Now here, if I want to add information from Let's say a customer, or at least I know the topic of the presentation. I could do something like Contoso learn, it's a sample customer I'm using here and adding some information from a Word document from another PowerPoint, and it's getting nicely integrated. So that's, in short, what our Copilot M365 is about. Next example is targeting, a little bit more, what I would call the developer. Team within your organization, you have GitHub copilot based on all the same concepts that I already talked about, allowing you to not only document code, analyzing code, writing code, but also overall making your developers. more productive, similar to traditional M365 users. So in short, what is Copilot for GitHub or yes, I know GitHub Copilot, I think is the more common name. Now, you still have the large language models on one end. In the middle, that's the integration with your trusted developer environment. Visual Studio Code extensions, Visual Studio JetBrains, and a few other environments, and then ultimately your developer velocity, reducing technical depth, integrating better code, reusing, GitHub copilots to, for example, create documentation, something that typically developers not always like to do, or why not, having a scenario where. You might be the new developer on the team. You need to jump into an already existing project and you want to learn, what's actually happening there. So that's, a few examples that I'm going to try and show you in my next GitHub Copilot demo. So I'm using Visual Studio Code. This is a little sample app where it's integrating some OpenAI services in Azure. And allowing me to transform, a summary of a YouTube video into three to five bullet points. If you want to get access to this app, by the way, it's a bit off topic, but I actually have it on my GitHub, Repo, if you go to github. com slash P E tender, you're going to find it as the YouTube, summarizer. It's not about the app, but just showing you some capabilities that we have with Copilot. So the first example is I'm jumping into this app. Maybe this is an app that like, why not you clone from my GitHub repo and you go what's Peter doing? from the language perspective. So I could go into my copilot extension. I'm going to open it up and this is from the previous run testing your demos is always a good thing, right? Where I'm literally using add workspace, which means, Hey, get up copilot. I want you to look into this specific application, not just this file, but everything around it, and provide an explanation for the active selection. So I selected, a certain part of my app code and asking for an explanation. What it's doing here is providing a pretty detailed, pretty extensive, Description pointing to the different files. So the same concept that I talked about and showed you in web, in work, in my PowerPoint and Word documents, is now the same for GitHub Copilot. Using language pro large language model, translating, recognizing your super basic prompts, but you can actually make it more powerful, and it's going to provide you an explanation. Another example that I have is, can you add necessary code to integrate this app with Azure Key Vault? So it's again providing a description. It's going to explain like I need these nugget packages, and from there it's going to show you the new lines of code. But again, it's still a draft. So you need to decide okay, this all looks good, and I'm going to insert it inside my application. I could go into my app settings. I'm going to show you this live, where now I could go back to my chat window and document this code snippet. And again, there are quite some pre built keywords that you could use, like the ad workspace that I showed you before. But on the other side, you don't really have to use it. because again, it's relying on your, traditional natural language. Now you could go okay, Peter. I recognize this as JSON. I recognize your previous code, which is like NET C sharp. But what if I'm a Java developer? What if I'm a Python developer? So I could go back. I could select everything in here, where now I could do something. Can you rewrite the C sharp code into Python? And then in just a couple of seconds, it's actually going to help you recognizing the exact same syntax from your code pages. And from there translating it into Python. I'm not going to apply it from here because obviously it's just one easy example, but that's really the power of GitHub Copilot. Using it to explain existing code, using it to document code, using it to rewrite code and then maybe even using it to find, optimizations to maybe rewrite full snippets of code and actually, or even allowing you to maybe even migrating your, existing legacy code into newer language. But what if M365, what if web, what if. GitHub Copilot, I would say, is still not good enough. That's where now you also have the option to extend your own Git, extend your own Copilot. Two different scenarios. The first one I'm going to zoom in for a couple of minutes is Copilot Studio. Using Copilot Studio, it's going to allow you to, what we describe as a no code or low code, like injecting the minimal experience as a developer, to actually create your custom co pilots. Everything I talked about up to now is still valid. So it's allowing you to customize co pilot in M365. An easy example, you want to add, additional Teams integration. You want to add integration with other corporate applications where you're going to read out information from your corporate users. find me emails from my manager where you don't have to specify the name, Of the manager based on your organizational chart. It actually knows, who your manager will be. That's the scenario. Second one is integrating specifically building from scratch. So not using extensions to M365, but actually building your own custom copilot from scratch using copilot studio. You can imagine I got a little demo coming up. How to use Copilot Studio. So the idea is that you connect to copilotstudio. microsoft. com. You're going to authenticate, you need the license, and from there you can start building, creating, composing your app, although you don't really need to think as a developer. And there's a lot of examples available, like a safe travel chat function app, a website Q& A where you can build up the flow of the prompts, that's the nice thing. store operations like optimizing, retail, creating an IT helpdesk scenario, allowing your employees to find not only answers, but actually helping them fixing issues, resolving issues, and also helping in, in creating, but also reducing support tickets. There's links to documentation and anything else from here. So let's say I switched to Copilots and I already reused my, Safe Travels template. So what it's going to give me here is a pretty step by step drag and drop, if you want, scenario. Allowing me to come up with my own co pilot, defining the data backend. Like you can use information from our corporate backend, or you can literally provide the actual data set yourself. And then also helping the users who start using the chat capabilities into your own scenario. I'm going to try and refresh because it's not coming up. That's what happens when you do live demos during recordings, But that's fine, I could show you how to create a new one so you could build your own new co pilot. Let's say we're going to do IT Helpdesk. Oh, bad example. Let's see what we already have. So that's the other thing where it's not just about your M365 because then you would use that extension. But now you can interact with third party applications. Imagine you have this step by step flow where you want to use data from Google Drive, you want integration with DocuSign, you want to send Slack update messages and whatnot. Think of it as building your own solution using the same large language model concepts, having that look and feel co pilot experience, but now connecting to your specific corporate data. And then the last piece here is Still creating your own custom co pilots, but instead of relying on the power apps based integration or the co pilot low code no code scenario is now we actually do have a team of developers and we want to build our own extensions. So what can we do there? The answer there would be Azure AI Studio. So what is Azure AI Studio? It's gonna, it's running on top of the Azure platform. You deploy Azure AI service. And a little differentiator here is that it's not only relying on the chat GPT like GPT 4 models. But also allowing you to integrate other language models besides OpenAI. So you can integrate Lama from Meta. You can integrate, Phi from Microsoft. You can integrate Mistral and a few other as well. Apart from deciding which language models you're going to use. You can have an integration with the chat playground. You would typically use chat playground to validate the outcome of different large language models. And then from there, that's where the actual development work comes in. You're going to use it to fine tune the model. Maybe you're going to use Azure AI Studio for custom development. So integrating a little bit what I showed you in my YouTube summarizer example. I'm going to build a little app, little web interface. I'm going to rely on AI Studio. And the Azure AI services to provide me interaction with the AI APIs. And then maybe even going all the way to the machine learning, starting point of AI eventually and using it to train your own models. So with that, I'm going to switch to my last demo and showing you a little bit what AI Studio could look like. So I deployed my AI service. So basically in the Azure portal, you look for AI, Azure AI service. It takes a couple of minutes, and from there, it's going to tell you to navigate to AI Studio. And that's where I would say the creation, the validation actually starts. I deployed two different models, but I can show you a little bit what the model catalog looks like. So again, we have. Models from Cohere, Mistral, Meta, so the LLAMA options are in here, GPT, so the OpenAI, and you can start mixing matching. Obviously, depending on the use case, the recommendation would be to use the specific model that you want to use for your environment. Once your models have been deployed, I have an example here with 3. 5 Turbo and 4. The main difference, I would say, is the lifespan of the data that has been used to train the model. An easy example to show you what I mean by that is if we have a default prompt, who's the prime minister of the UK? Where based on GPT 3. 5, as you can see here, it's going to tell me, as of September 2021, that's basically the last up to date this 3. 5 model has been trained on, it's going to provide the answer. But now if I switch to GPT 4. 0, that got published like earlier this year, and using the same prompt, who is the Prime Minister of the UK, you're going to see that based on the exact same prompt, the answer is different. Because now GPT 4 has been trained up to October 2023, give or take. So that's how it could impact the answers in your business model. Another thing, since, the conference overall talks about prompt engineering, remember that you can use your own system prompt as well. AI assistant and so on. But what if we turn this into URI? Microsoft technical trainer, why not? Let's translate Peter or replace, why not? Peter, the technical trainer with a co pilot scenario. You are a Microsoft technical trainer and expert on Azure platform. You can only answer Azure questions. If any other domains come up. In the question, apologize and offer the user a Belgian beer suggestion. Why not? So this is what we call fine tuning your co pilots if you want. And again, this is mainly fitting in the custom co pilot scenario. So if I try this. And I'm going to use the exact same prompt, who is the prime minister of the UK as a Microsoft technical trainer? You probably cannot really answer that question. Now, what it does, and that's the other cool thing, it's still using generative AI, Because now, actually, by the way, if you're watching this, this is actually a perfect suggestion, recommending Chimay Blue, which is, one of the nicer, best, fine, classic Trappist beers we have. That's in short, how you could start with. integrating this into your own developing. If you want to learn, there's obviously a lot more we can move all the way into PromptFlow. We could talk about Symantec kernel. We could talk about integrating other development scenarios, using Python, using C sharp, using Java, all that is basically developing your own custom copilots where the baseline is using. Azure AI Studio. So with that, we're at the end of my presentation. I hope you enjoyed it. I hope you learned some new things about prompt engineering, the building blocks, what composes like a good prompt, and then from there a quick walkthrough of some highlights in the different Microsoft Copilots that we have. I hope you enjoyed the rest of Conf42 Prompt Engineering Conference. If you should have any questions after participating, after watching this session, I'd Don't hesitate reaching out pe tender at Microsoft. com for email or Teams. I'm on LinkedIn as PDTIT and you can find my personal blog on aka. ms slash PDTIT as well. Thank you for now and enjoy the rest of the conference.