Transcripts
1. Introduction: What if I told you
you could build chatbots, virtual agents, AI assistance without even
writing a single line of code? In this current class,
I'll be teaching you how to use power virtual
agents as part of the Microsoft suite of Power Platform tools to
help you create chatbots, virtual agents, AI assistance without even writing a
single line of code. Yes, you heard me right. No
coding required whatsoever. And which I'm going to
teach you how to go through the process from A to Z such that by the
end of the class, you have a fully
deployed virtual agent, virtual assistant regardless
of your application. You can use it for
questions and answers, customer service,
responding to queries, automating tasks, or responses
for search purposes, and the list is endless. This is very important
in today's world, in today's market where artificial intelligence is actually disrupting
many industries, and you need to be
up to the game, having those skills, having those expertise in
order to stand out. And I'm going to
show you how to go about this from A to Z without writing a
single line of code in the best way and the
fastest way possible.
2. Your Project: Your project for
the class revolves around using power
virtual agents to create your own chat bot for
your own unique application, whether for marketing,
questions and answers, customer service, maintaining
your branding image. The list is endless. The application is endless, feel free to get innovative
with your project. After which you're
going to be sharing your virtual agent with the rest of the
community for feedback.
3. Power Virtual Agents Fundamentals: Back. In this lesson, we are going to learn about a very powerful tool
within the Power platform, which is the power
virtual agents, which is another
word, a chat bot. Now there have been emerge from the normal website of Power
Virtual agents to copilot, in which we have the
ability to actually create virtual
agents or chatbots, which are customizable
to various businesses, websites, and operations easily involving low code or no code. So Power Virtual agents
is a low code or no co platform within
Microsoft Power platforms, which enables you
as a user to create AI powered chatbots without the need for advanced
programming or any skills. And the great thing about
them is that these chat bots, they could interact with
customers, internal teams. They could automate
conversations and provide real time assistance in addition to actually conducting
automatic tasks. So you have the
ability to actually create a virtual agent, a virtual chatbot, and add it to your website or multiple
websites and to train it. This is very powerful as we're going to see once we go about the hands on building
of a virtual agent. You have the ability to fully customize the agent
based on your needs, whether for a travel company, whether for your own school
work, whatever it is. You have the ability to
build these chat bots, which are customizable,
trainable, and they provide you with solid response based
on your requirements, saving you time,
saving you effort, and making sure that you eliminate a lot of back
and forth communication. All of this is
going to be covered through the power
virtual agents.
4. Advantages of Power Virtual Agents: So what is the great thing
about power virtual agents? First of all, they provide you
with the ability to create customized chat bots that have the ability to communicate, respond, and ask questions, provide information using
no code whatsoever. So it's a no code bot creation. It has the ability to integrate with other power
automated tools, for example, power apps, Power BI, power automate. All of them, they could
be integrated with power virtual agents to
provide a holistic solution. It's an AI powered conservation chatbot,
conversations chatbot. It means it has the
ability to provide customized responses based
on certain prompts or input. And the beauty about it
is it's customizable. So you have the ability to
actually train the chat bot, provide information
and contacts, to provide specific answers
based on your contacts. And finally, it has
multi channel support, whether on an application. Or a website or a
blog, wherever it is, you have the ability upon
completing once you have done you have fulfilled and completed the creation
of the virtual bot, you have the ability
to actually add it and inject it to any or
to many channels, such as websites, blogs, apps, and on multiple locations. That way you have the
ability to actually provide the same support at
various locations, saving you time,
saving you effort, saving you the whole hassle of back and forth communication
and allocating information. All of these things are
going to be executed with a virtual agent
or a chat bot, which is AI driven, AI trained. That way you have the ability to think outside the box and create a fully integrated
solution from A to Z.
5. Virtual Agents Access and Interface: Welcome to this lecture in
which we are going to explore the power of virtual agents within the Microsoft
Power platform. First of all, how do we
get to the virtual agents, which is considered to be
a chat bot, sort of speak? So if you go to the
navigation over here, you're going to put in
copilot.miicrosoft.com to land at the home page. Now, you have to be
careful with this. I've already logged into
the virtual agents, keeping in mind that there's
a combination now between copilot and virtual agents,
power virtual agents. It used to be
dedicated by itself, and now it's combined as
part of the copilot studio. So this is often an
area of confusion. Where you have
students looking for the virtual agents and they
run into the copilot studio. So just simply put in the
browser copilot studio domicrosoft.com in order
to land at the homepage, which basically looks like this. And it will have the
following display which says, describe your copilot
to create it. As you can see, the
initiative now is to combine the virtual agents and copilot together such that it's now under the
umbrella of co pilots. And within the integration
of artificial intelligence, you have the ability to now combine the power of
artificial intelligence to create bots virtual agents to help you out with various tasks within your day to day business. So that being said, also, we need to keep in mind when
you are trying to log into the copilot studio or
the virtual agents, if you have been using
a certain account for Power Apps, Power BI, Power Automate, you
might notice it will not work if you go to the log
in to the copilot studio. Why? Because copilot
Studio requires the administrative
account to log in, not just simply a
testing account. So you need to go to the
original environment account that you used to set
up that account that you have used to set up the Power platform at the beginning. The main account to set
up the Power platform, you need to use that
in order to access the copilot Studio or
the Virtual Agent. Keep that in mind
because it gets quite confusing when you
jump from Power Apps, Power BI, Power
Automate Power pages, all of them, the work with
the same login details. However, when you go to the virtual agents,
power virtual agents, you notice that you cannot
have the access to do so because the virtual bots are integrated at
administrative level. So they require higher
level of access. So that's why you need to use the main account that you have used during the setup phase. So that being said,
it's very important to highlight this now in
this current lecture, we are going simply to
explore the interface of the copilot studio or the virtual agents within
the power platform. Now they're combined together. This is the homepage where
you have the main display to use prompts to
create chat bots. Then you have any chat bots
that you have created. You have templates and
learning resources. Then you have the create option where you have the ability
once you click on, we're going to leave
it to the last to start to create
your virtual agent. Once you click on copilot, this will show you all of the
bots that you have created. For example, I created
a travel vista bot. This is something
that you could use to help you with
travel, for example, and you could create a
new copilot directly or import a copilot if you
have created one and saved it. As for the library,
this includes a set of applications
and integrations. You have the ability
to add one in order to create some
sort of connections. Since you are building a
chat bot or a virtual agent, or a copilot butt. Sometimes you might need
what we call as connectors. You need to get data
from somewhere. You need to integrate
this somewhere. And once you go to the library, you have the ability to
find these connections. If you can't find it, you can add it as an item over here. Now, let's transition
to the important part. To help us get up and running. You click on Create, you
will have the display now to create your
virtual agent. As you can see over here, we got two different options which we are going to
be discussing up next.
6. Methods to Build a Virtual Agent: This current lesson,
we are going to take a look at the methods
that you could use to help you create your virtual agent or what's
called Do the copilot. So once you go to
create over here, you have the following option, new copilot, new
Microsoft copilot action. Now, what's the difference? So the first method for
creating a copilot, it leverages end to
end generative AI. It means going to be powered
by artificial intelligence from A TZ for completing tasks. Sharing knowledge, answering questions, getting analytics. So in other words,
you're not going to be adding raw data or getting some connection sources
from your company or getting some data
from your company that you're going to
be integrating this. No, you're going to use mainly artificial intelligence using accessible data sources to help you build your power bot
or the Power platform bot, as we are going to do within
the hands on application, which is the easiest option
to get started with. The other option is to
go for what we call as the copilot action
in which you build a customizable but
or Microsoft copilot for your organization.
This is advanced. This is a subject for
a later discussion, but we're going to explore the differences for you
to have a clear idea. And the third option is
you start with a template. So let's explore the first. If I click on New copilot, is going to prompt
me to this page, which we are going
to be discussing with details. Let's go back. However, if I go to the new
Microsoft copilot action, you will notice I have
the following display. I get connector, conversational,
flow and prompt. These are different methods
to develop your butt. This is very important and
often it gets quite confusing. For example, when
you are creating a copilot or virtual agent
with a connector method, it means you are going to
get external data sources. Based on a prompt, you
are going to describe it, give some description and
form of AI in order to build that chat bot that
copilot conversational, this is going to be developed
with topics and responses. Think about it as an
automatic responder. Whenever you go on a website, you ask questions, you find
answers automatically. This is a conversational
virtual agent. And the fourth one,
which is the flow, this is where you integrate
it with Power Automate. So you could use Power
Automate to actually create a flow where you
have a virtual agent. Now, these the flow and the connector are for
advanced practices, since they need organizational resources or sources of data, and they need to be integrated
with the other power apps, for example, tools, Power automated tools,
Power pages, et cetera. So it's beyond the
current application, which is helping us focus on the fundamentals
and to get up and running and getting up to
speed with the learning curve. And the prompt is quite easy where you give
it a prompt to help us design virtual agent. And finally, the conversational, we're going to
tackle this as well to help us create a
virtual agent which is focused on communication,
topics, and responses. So you get the
differences right now. When you are trying to build a virtual copilot
or virtual agent, either you have the
option to build one using artificial
intelligence from AT Z, or you could go to the new
Microsoft copilot action, which gives you multiple
options to integrate based on connect
resources or flows, or you could use prompt
or conversational topics. That way, if you have a specific application
within your company, you could go for
the new copilot. And for the ease of practice, you can just simply start
things off with a template. So that being said, in
this current lesson, we have went through the
various methods that you could use to create a
virtual agent or a copilot. Now, within this current
segment of the course, we are going to focus on
the new pilot methodology. We're going to test a template, and then we are going to have a prompt design chat
bot or virtual agent, conversational
chat bot, as well, we're going to design that one. However, for the
connectors and flows, we're going to leave
this for future courses, integrations, and updates
since this is a bit advanced. So that being said, now you have a clear idea about
the interface, the different methods
that we could use to help us create a virtual agent, and now we are ready to dive in to build our virtual agent.
7. Creating Your First Virtual Agent: And welcome back. So
the first thing we're going to do is we are going to create a virtual agent
using a template. So the first thing
you need to do if you're not familiar
with this right now, you go to Create
you click on it. You're going to land
on the following page. And what you are going
to do right now, you're going to select
any of those templates. Keeping in mind, these
templates are pre built. As you noticed, some of
them, they are coming soon. Some of them are very powerful. You got CV match,
you got job craft. This is very, very powerful. You got coming soon
for approvals manager. So these are powerful
virtual ages that you could use once
they get released to help you automate many tasks. Let's start with the safe
travel, for example, template. Here we go. So this template has already been set up for us. You have the option, once you transition to this page,
to give it a name. So I'm going to
call it safety Bd. Okay. Safety bud. Here we go. You can't change the
icon. It's up to you. Then you are going to give some descriptions where you are going to add some context to it. Why? Because once we integrate artificial
intelligence in this, think about it like a prompt. It provides answers to
common travel questions and related health and
safety guidelines and provides safety suggestions
for passengers. So I'm giving it the context. This is the task,
what it needs to do. Instructions. So how this
virtual agent should behave. You are an assistant that helps answer questions
about traveling. Safe both domestically
and internationally. Please respond in
a polite manner. Make sure that your tone, for example, is friendly. This is part of prompt
engineering, by the way, where you add some context in order to find
tune the result and then include
professional details, not just simply recommendations. Here we go. And this
is the beautiful part. If you take a look at
this, Add knowledge. You have the ability
to train your bot. You have the ability to
actually add resources in order to allow it to train and learn in order to
respond effectively. For example, once you
click on the icon, Add knowledge, you have
this prompt window. Add knowledge resources. So you have the featured
options, which are websites. You can just simply
add any website, and the virtual agent will
get the data from it. You could have a SharePoint
list from OneDrive. These are coming soon, by
the way, Microsoft fabric. You could upload a file. Now, some of them, they
are not available. To be used, and some
of them require for you to go for a premium version, not a trial version to
actually make it work. You have the data verse. If you're going to
integrate within your company and you do
have other connections, you could connect this
to your enterprise data, and currently it's on preview. So you have, for example, file share, CSV files. All of these are different
sources of information. So for the sake of simplicity, I'm going to click
on public website. So how do we go about a website? Simply select any website which
follows these conditions? It should be indexed or
found by Bang, for example, because Microsoft
is integrated with Bang don't use sites with forms or comment
from end users, which can reduce the
relevancy of the answers. Do not include query strings, more than two levels of depth. So basically the URL
should be quite generic. Let's go for ww.tripadvisor.com,
for example. Here you go. So you
add any website you have in mind
based on the theme, and this is quite acceptable. Got the description for it. Let's add the website.
So again now, I got two websites I'm
going to use to help me train my safety
bud virtual agent. You can also edit the
language, by the way. Now, let's click on Create. And this is the prom window. We're setting up the copilot, and now our copilot
is going to be up and ready as a safety virtual
agent for travel. So this is the copilot. Any virtual agent
that you create is going to be displayed over here. Now, we are going to examine the various settings
of our virtual agent.
8. Creating A Virtual Agent from Scratch: Welcome back. Now we're going
to actually learn how to create our own copilot or virtual agent
completely from scratch. So we're going to navigate
to a new copilot. Let's give it a name,
and we're going to have a copilot to help us
with food recipes. Okay? So food buddy. Here we go. So this
copilot will help us. This copilot, or I will help you generate food recipes based on a certain
cuisine. Here we go. That's it. And
then instructions. You will be providing a list of recipes based on certain
cuisines as an expert chef. Make sure that you
are friendly and include all necessary
ingredients and details. So what I'm going
to do right now, I'm going to actually
create a food body a virtual agent to
help us generate food recipes based on
a certain cuisine, for example, including
all of the ingredients. So let's add some
knowledge sources, a public website, w.cooking.com. Another option would
be www.food.com. Whatever website
that you have in mind that you could use
to actually help you come up with recipes,
wt, food recipes.com. I'm just simply using these as random sources. I'm
going to add them. Now I'm going to
click on Create. Now our co pilot will be set up. Now I'm going to
test it. Let's see. Let's take a look at the
end result. Here we go. Hello, I'm food buddy,
a virtual assistant. Just so you are aware, I
sometimes use a I, okay? If you provide a website
during creation, try to ask me about it
and give it some context. So we've added some knowledge. We give some resources. It just simply has basic topics, no actions, not
being published yet. I could add conversations to it as we're going
to see just a bit. However, let's test it out. I want to cook an Italian dish from the Italian cuisine,
for example. Let's see if it's able to actually come up with
a proper answer. Here we go. For an Italian dish, you can try making
spaghetti la carbonara. Yes. This Roman pasta
dish has been excellent. So it's up and
running. So it gives me actually the ingredients. So how about now let's
add a topic to it. I'm going to add a topic. Let's try to build this
topic using copilot. Now, cooking steps.
Cooking steps. After the user picks
the name of the dish, create a list of steps
to cook the dish. So now, once the user gives
it a name for a dish, for example, we're going to come up with steps to
create the dish. So let's see how
does that look like? First of all, phrases, how
to cook, steps to cook. So this is what you call as
the topic. This is a trigger. So if I click Edit,
this includes all of the things that
you might write over here to actually ask the chat bot or give
me the steps to cook. So you add every single context that you could use to
help you with that. So this is a trigger word. Once you type this over here, now you have the ability
to trigger this. So what is the question? What
is the name of the dish? Use entire responses. It could be any
response whatsoever. Let's take a look
at this. We could search for other valuables. We have the dish name.
It's already included. We're not going into coding, so there's nothing to
be concerned about. Then we have a message. After we trigger the question, here are the steps. We could add extra items to it, but let's test it
out. Here we go. How to cook, for example, I used the phrase from the display phrases
to trigger this. What is the name of the dish? So here you go. How to cook. Then we have the
name of the dish. I'm going to go for
let's use this name. Here we go. I'm going to
here we go copy this, and just simply
paste it over here. I gave it the name of the dish. Here are the steps to cook,
Spaghetti ala carbonara. Prepare the ingredients, follow
the recipe instructions, cook the dish according
to the recipe. But as you can see, I don't
have the detailed recipe. Why? Because I did
not train it enough. So what I'm going
to do right now, let me try to edit this. Here are the steps to cook. Prepare the ingredients,
follow the instructions. Now, I would like to
add something else. This does not work
properly for me, so I need to remove this, and I'm going to
delete this, actually. I would like to have
a proper response, like the actual steps. So I'm going to click on this. Send the message, ask
with adaptive card, add a condition, call an action, or you have advanced features. So I'm going to go for
generative answers. This is where I'm going
to have artificial intelligence to actually give me not just something a topic which I have added
and filled by myself, but include the flexibility to come up with a
generative answer. So now I'm walking you through a hands on application to see how I'm building
this virtual agent. So I've added some knowledge. Now I've added some
topics using copilot. I'm trying to modify this. It all starts with a trigger, and then we have
any other actions which follows that trigger.
We got a question. We could modify this.
I kept it as is. But I've noticed that the answer I'm getting is quite generic, just simply another statement. I need something generative. So I'm going to click
on the Plus button. We could send a message,
ask another question, add conditions, ask with a card. You could add images as well. You could have the ability
for variable management. You could just simply
change the variables. You could have topic
management where you can just simply go from
one topic to another. Or end the current topic. Once the discussion goes
to a certain point, you can just simply end the discussion, end
the conversation. These are different options
that we could tinker with. Call an action, for example, we have creative flow,
create a prompt, or skill or based action. These are integrations with artificial intelligence,
power automate, and additional skills
or extensions that we could add if we are trying to connect it
to an external end. Advanced option, we have
generative answers. We could just simply
send HTTP request, log a custom telemetry
event, send an event, send an activity,
authentic these are different criterias that you could implement based
on your specific cases. Like I mentioned, we're
keeping things quite simple in this current course and in this current application
for the fundamentals, for the essentials, for
the beginner's level. But the subject is extensive. It's quite endless in terms of the things that you
could learn and cover. And this is something
that we're going to be leaving for advanced courses. So now to get
generative answers, we're going to click Generative
answers. Here we go. So the data source
now enter the value. I'm going to have the dish name. This is the value which
I'm going to actually use. The data sources, we've added
different data sources, such as the previous ones. We could add a knowledge source. I'm not going to add extra stuff because I'm going to use
artificial intelligence. We could search for content
online using Open AI, where I can just simply click Get search results
from knowledge sources using Select mode. I add knowledge or allow AI to use its own general
knowledge preview. And that's simply it. And now we are going to utilize this. Here we go. Classic
data, click on Save. Now we'll notice things
will be quite different. So what I'm going
to do right now, the answer will be a
generative AI response. So I'm going to create this
and let's paste this now. See the difference. Now you
have the ability to integrate different options
within your application or within your virtual agent. Here we go. So to cook spaghetti la care we I want to follow these
general steps. Cook the pasta according to package instructions in a pan, cook the panzera
bacon until crispy, reserve some pasta water, then add it to the
pan with the panzera, toss in the excellent. And it got you the references. So as you can see, I'm
building my copilot. So now it's quite ready. Why? Because I can just
simply give a recipe. I want to cook a dish, and then I could pick a recipe how to cook,
give it a dish name. I will come up with all of the steps to do it,
and that's mainly it. So this is a very
basic application to help you get up and running. With creating your copilot. So now I'm going to click on
Test. Let's test this out. We're not going to
have any problems. It's working perfectly fine. Topic checker, there are no
errors in case of any errors. Once you click on actions,
like I've mentioned, this is quite advanced
because you're going to integrate this with power
automate and other features. And we're focusing on in
this current course on helping you build your virtual agent for
your own practice. So we have created a
safety body virtual agent Using a template, and
we've created a food body, a virtual agent which
is quite from scratch. If we click on overview, this is the name. We
could change this. We could change the icon.
We have the description. You could add the knowledge. You could add a topic.
There's no need to add anything manually
or code based. You can just simply
click Add topic and use copilot to help
you build that topic. And then finally at
advanced stages, you have the ability
to add actions. And you can add extra knowledge to further fine
tune the results. We've seen how I was
able to actually modify certain topics in
order to get certain results. And once you click on a certain
topic, you have the idea, you have the ability
to get ideas based on trigger words which are
going to be added over here, and then you could add different
actions as you please. Think about the logic that you would like to follow in order to engage with your users and the logic that we
have followed over here, phrases related to
cooking, how to cook, steps to cook, and
a guide to cooking. Then what is the name of the
dish? It was a question. Then instead of having
just one answer, we're going to create
a generative answer using artificial intelligence
by clicking on the plus, going to the advanced
generative answers, and to get better
and better answers, you need to integrate this with additional knowledge sources. So now we have a
complete copilot which is ready to be deployed. And when I simply click Publish, I'm going to be
publishing my co pilot, which is quite ready right now. We have built it from scratch. We've added some
extra functionalities by modifying the topics a bit without going into coding and without
integrating external actions, which could be added at a later stage for
advanced virtual agents. But for the sake of simplicity, for you to get the idea
to create your bots, you have multiple options. We created one from scratch. We created one using a template. We used copilot to modify them, and now our food body pilot is ready simply when
I click on Publish, and if you have the
ability as well, to modify the settings,
just keep that in mind. You have the ability
to modify them in different ways, based
on your preferences. For example, you can go
for the classic option, which means every topic gets a trigger and
an action based on certain words that you
add or use generative AI, which I've added within my topic to actually use the power of artificial intelligence to help us come up with responses. So these are very, very important modifications
to keep in mind. Now, we have the overview. We got the knowledge in check. We got the topics in check. We got the actions
for advanced stages. Once we see that, we need to add external connections,
we can go for this. Then you got the analytics. If anyone's using the copilot on your website or
your application, you're able to see this through integration with Power BI. And finally, you have channels where you get to
publish your copilot. Now, our copilot is published, but I didn't connect it
to any of the channels. If you click on Microsoft
Teams, for example, you'll have the
ability to connect the Team's application
to your copilot. If you got a custom website, you have the ability to connect your custom website to your copilot or the
mobile application. So it's quite straightforward. We started from scratch,
and we went all the way in terms of publishing
our copilot.
9. Exploring Additional Settings: And welcome back. So
now we've created our safety bud virtual
agent using a template. Now, when you're
using a template, you have to keep in mind, there are many things which
have been pre built for you. So if you take a look over here, for example, this is the
preview of the template. It tells you that your
co pilot is ready. This is the interface where
you get to test your copilot. Hello, I'm safety Bud,
a virtual assistant. It gives you a briefing
about the virtual agent, and now you're able to
directly engage with it. Now, let's take a look
at the description. You have the ability to edit the details of your
virtual agent. You're able to add and
remove knowledge sources, allow the integration of
artificial intelligence or not. This is where it
gets interesting. You have the ability to
add topics and you have the ability to add actions.
This is very important. When you say add topic, it means you have the ability
to create a conversation. Now, since it's a template, there are things which
have been pre built. Let's take a look at the top. For the knowledge
sources, they are ready. Simply you could add
self explanatory. You could add
knowledge sources to further enhance the ability
of your virtual agent. For the topics, you have
the ability to create topics which are
basically blank form, or you can just simply
describe it with copilot. If you don't want to code, you don't want to do anything related to actually
building this from scratch, you can use copilot
to help you come up with these
conversational topics. So what do they mean? If I click on any of
them, for example, this will transfer it into
a form of a trigger and an action based interface where you have copilot to assist
you in the process. You have the ability
to check the topic by clicking on this topic checker in case there are any errors. You have the ability to take a look at the analytics which integrates this with Power
BI as we're going to see. So a trigger, once a
conversation starts, you have the ability to edit the parameters of
a conversation. When I click something, this
will count as a trigger. Then a message will be prompt. Hello, I'm so and so, which are basically
the details over here. Notice this includes
a bit of coding. You have the ability to modify
this when I delete this, for example, be
reflected over here. So you have the ability
to actually modify the prebuilt settings.
This is very important. You have the ability to modify the prebuilt settings which
are within the topics. And every single topic
is considered to be a form of a conversation
where you have conditions. If something happens, do this, if something happens, do that. So all of these topics have been prepopulated in the template. So we're not going to be
building this template from scratch because many of these topics have
been built for us. For example, multiple
topics match. Have a trigger,
you set variables. Again, this requires a bit of coding, which
like I've mentioned, we're not going to
be diving into, but it's good to have the logic behind it to understand
how it works. And let's say you would
like to enable or disable any topic you have
the ability to do so. But what if you would like to add a topic but you
don't want to code? Simply click at topic, create from the
description with copilot. The copilot prompt
would open over here, and it will tell you just simply to give it a name and
what you'd like to do, and then it'll
create it for you. Let me show you how.
Travel Inquiry. So ask the user about their preferred
travel destination at the beginning of
the conversation, for example, conversation. And then this will
create a set of conditions to actually
come up with a topic. So it's building a topic, as we can see, here we go. So now this is a trigger. Describe what the topic
does. Here we go. What is your preferred
travel destination, user response, and that's
mail it. So these are good. So you have the ability
to actually come up and tinker with us using different ideas and
different contexts. Like I've mentioned,
at this current stage, we're just simply browsing
through a ready made template. We're not building it, but
we're examining how it works. So now we have an
overview for our copilot, the description, the knowledge
base, some key topics. Now, for advanced users or
for advanced practices, you have the ability to
actually integrate tasks. Once you click on add action
you have the ability to create connectors or connect it to flows within
Power Automate. We're not going to
cover this because this is quite advanced in which you have to
set up Power Automate, and then you're going to be connecting it to
the virtual agent. For example, if a
user submits a name, you have the ability
to get a worksheet or get a row or run
a certain script, which are endless in terms
of their application. Like I mentioned, the domain
that we have is quite broad with extensive numerous
areas of application. But for the new users who would like to get up and running
with virtual agents, we're not going to focus on the highly advanced stuff
such as adding actions or tinkering with topics by adding conditions which require a bit of code involvement. We're going to
bypass this by using copilot to help us
building this at the same time using either ready made templates or using a basic setup
to get the job done. And finally, once we are
done with everything, we have the ability to publish our virtual agents
we're going to see. So we covered overview, we covered the knowledge, topics, actions, and finally,
you have the analytics. You can take a look at any data which has been collected by the virtual agent if it
adds data to your database, and that will be displayed in the form of a
report on Power BI. And finally, you
have the channels. Once you publish
your virtual agent, it will be deployed to channels. For example, you can
add to your website. You could add it to your
mobile app to Microsoft teams. So these are either
accessible or require that you go to Again, the premium paid version of
the Microsoft Power platform. Now, that being said, now we
have a full rounded picture. Let's test our copilot. So this is simply the template, and we're keeping
things quite simple. I would like to travel to Italy. Now,
let's see what it does. Now it's going to actually
browse for information, and it's going to be sharing
with me some updates. Take a look at this. What is your preferred
travel destination? Notice, this is the topic
that I have added, right? It popped at the beginning
because I gave it the prompt. So instead of going
doing this by myself, into coding, building the code, adding the conditions,
why waste time? Go for the autopilot
copilot assistance. It will help you
get the job done, and you're able
to see the topic, but you have to be solid
in terms of using prompts. I'm going to type
Italy. Here we go. Now, it's going to actually
apply the topic triggers and the conditions to
actually help guide my input. Now, that being said, you have the ability
to tinker with this to keep to see
their responses, if it's actually in line with
your expectations or not. What is your preferred
travel destination in Italy? I want to create a travel plan. Here we go to Italy. So I gave it the destination
at the beginning, instead of just simply
jumping into the description. Now, it's going to
loop again and again. Notice, we created an error. Why? Because every time
I engage with this, it's going to ask me for the preferred
travel destination. So what I'm going to do
in this current case, I'm going to close it. Here we go. I turned
off this travel query. So once I've done so, it will not show again.
So here we go. Let's repeat the prompt.
And you test it out. So I'm walking you
through the steps of back and forth
communication that you would be expecting as
you are building your pilot. Take a look at this.
Escalating to live Asia is not currently
configured for this bot. However, this is where
the bot could provide information about how to get in touch with someone another way. Is there anything else
I can help you with? So it seems that the bot is not able to handle
creating travel plans. Why? Because we
created a safety bot. It hasn't been trained.
So you get the idea. Now, you integrate
knowledge to train it. You could add co
pilot for topics. I'm going to refine
the prompt what are the safety procedures
for traveling to Italy? Keeping in mind,
if the information is quite limited,
it will struggle. Why? Because I gave it
just simply two websites. Now, let's take a look at this. Now it will dig into
the knowledge base to actually get for me
a solid feedback. So what are the
safety procedures for traveling to
Italy? Here we go. The safety procedure
for traveling to Italy includes checking
the excellent. So this gives me
a solid response, even though it's quite basic because I did not
train it enough. I did not add topics the
way I would like it. So based on your case, you have the ability to
actually modify this. Let's say this is
quite satisfactory. So what I'm going to do right
now, I'm going to test it. You click Otext, it's done. Now I have the ability
to modify the settings, either taking a
look at the name, we browse the generative AI. How would you like it to
interact with people? You need to use AI to go for this or topics to go for this, such as phrases and action. This is sort of a
manual approach. This is using generative AI. You have the ability to modify the strictness of the
content moderation. Then you have the
security options. These are important things to
keep in mind, for example, authenticating the
user's identity, taking a look at
the web securities. These are bit advanced features. Then you have the
authoring tools, then you have the entities
with which you are going to be utilizing within your tables, which would be reflected as
well on your virtual agents. You could add skills. Now, skills extend your co pilots
conversation capabilities by automating series of
actions within a topic. So you could add certain
skills, for example, like booking an
appointment, email, managing tasks, which again, these are next level
advanced features that you could integrate within
your virtual agent, and then you have the ability to optimize for voice where
you have the ability to actually input
voice optimization if you're using phone call, for example, but this is again, depends on the application. You should have the ability
to integrate, for example, with empower automate phones, phone calls to actually do this. So as you can see,
you have the ability to tinker with us
on various cases, whether automation
through phone calls, through skills, through tasks. And then once we
are done with this, you simply click Publish and your copilot is
ready to be deployed. So once we're going to be
publishing their copilot, it will take some
time, first of all, to include all the details. Then you will have the
ability to see it over here as one of your deployed pilots. So that being said, now you have a generic idea how to go
about using a template, modifying it,
tweaking it up a bit, and then you are good to go. When you are publishing your
copilot or virtual agent, you have the ability to actually integrate it on your website,
on your applications. You have the ability to make it available or as part
of integration. For example, for channels, here we go. You
can publish this. Then you could click over here
to integrate it with them, for example, Microsoft Teams. If the demo website
was available, like I've mentioned, some of them are available for preview, and some of them they need to have authentications because it should be on a premium version
to be able to use them. So once we are done
with publishing, you can just simply
click on Demo website or mobile application
or custom website, you add the URL, you hit the
publish, and here you go. Now your copilot
will be ready to be integrated in all
of these channels, helping you communicate
with your customers, with your colleagues
based on the setup that you have made it to
do in the first place.
10. How to to Display Your Demo Virtual Agent : We've seen how to create virtual agents through
various approaches. And once we have said when we go about publishing
our virtual agent, we need to make sure
that either we have actually a premium version, or there's a different way that you could go about testing this through an
authentication process that I wanted to share with you. Let's go for the food body application that
we have created, and we navigate to publish. And once you click on publish you will notice that
it will take us to the window where we have on the channels where
we tend to deploy this. And the only thing which was available is Microsoft Teams. However, I'm going to share
with you away right now, in order for you to have access to all of these
deployment mechanisms, and you have the
ability to actually go on a demo website
to display this. So how would we go about this? Simply click on
the three buttons over here, go to settings, click on Security and then
go to authentication. And this will changes from
authenticate with Microsoft to no authentication
because once you keep this with authenticate
with Microsoft, it requires access to a certain account,
a premium account, a paid account within the Microsoft pool
in order to be able to publish it or test it with the need to
access an account. So if you get the idea,
you cannot test it if you do not have an actual
account to test it with. However, once you shift it
to no authentication needed, it means once it gets
deployed to your website, the user does not
have to be part of the Microsoft accounts pool, let's say, to test it out. So now the window
will change and you have the ability to
see a demo website, or you can see here
or a custom website. If you click on custom website, you can just simply copy and paste this code
into your website, and that's mainly it, and your virtual butt
is up or running. Let's take a look at
the demo website. And once we go for
a demo website, we have the ability to actually test it as if it's
on the live website. So now it's loading the demo web there's a different way to go about this, which
will save us time. Click on Demo website, and this will open an actual
demo website straight ahead. And here we go. We got
the food body in display. Keeping in mind, this
is a basic application. We did not tweak it extensively, just simply for
demonstration purposes to help you get the idea. So now we're going to give it some context from the
topics that we have added, including the
including what we have to keep in mind the
output from generated VI. How to cook, for example, is going to ask me for
the dish name, right? So how to cook. This is the conversation that
we have in mind. It will tell me which dish. What is the name of the dish? I'm going to say
spaghetti. Here we go. So now, it's going
to actually give me the steps to cook it based on the knowledge pool
that I've provided. So now you get the idea. We've created a
bot from scratch, virtual agent for food recipes, we give it some knowledge basis, some tools, and here we go. It's fully functioning
up and running. It's giving me different
recipes to cook spaghetti, different options. Here we go. And even the resources
and the references, and the more knowledge
sources that you add and the more topic tweaks
that you could incorporate, the better the output will be. So here we go. So we've created our virtual agent from scratch, and we have deployed
it successfully.
11. Building a Virtual Agent using AI: Welcome back. Now
we're going to go about creating a copilot
not from scratch, but using artificial
intelligence through prompts. Now, for the
connector, the flow, and the conversational,
we're going to leave them for
advanced practices. However, in the future,
we might integrate conversational for the advanced
users within the course. But at this current stage, prompt is also essential
to learn about. We're going to click on Prompt, and it's going to have
the following window to guide me through
the steps I need to create my virtual agent using
solely prompt engineering. So what is the action name?
Provide your action details and select on which solution
you wish to save it. Now, this is very important. You need to have
solutions in place. If you have learned about
Power Apps, Power BI, Power Automate when
you create a solution, it's like a package,
which includes data. And then if you have a
certain solution in mind, you're able to integrate
it with the virtual agent. If not, you're going
to keep it as blank, and it will create a
new solution for you. Action name, let's say, we're going to have a financial advisor,
financial advisor. Into the prompt
action description, the virtual or let's speak to
it as directly as possible. You are an expert
financial advisor, and you will be providing
general financial advice to users in a professional and
friendly friendly tone. Here we go. Very simple. Click Next. Then we're
going to create a prompt. This is where we integrate prompt engineering
best practice. I'm going to briefly
walk you through some powerful elements that
you could incorporate. So we're going to
define our intent. What are we trying to do? Summarize a text,
extract information, classify, analyze, respond to a complaint. So
what is the prompt? You will be providing
financial advice as an expert financial advisor based on inputs from the user. Now, let's take a look at
this. Test the prompt. So now we're going to actually test the prompt and we have the ability to add certain inputs and data
to actually modify those. Let's take a look
at the response that we have gotten so far. And you have the ability
to add input data. You could add extra
information to help you inject the data inside the prompt to make it
even more fine tuned. So all inputs and data
are automatically used to answer plus insert to
fine tune the prompt. You can see over here. So
I appreciate your trust, but I must clarify that I'm
not a financial advisor. It's important to consult the
qualified financial advisor for personalized
financial advice, tailored to specific
situation and goals. They can provide guidance,
budgeting, retirement. Okay. If you have any
general questions, feel free to ask. So as you can see,
just to make sure that the response
is quite authentic, the artificial intelligence
is not able to act as an official
financial advisor, but it will share some advice. So what I'm going
to do is, let's change this. Here we go. You'll be providing as an
expert or instead of an expert, as a well informed advisor. Here we go. Based on
the from the user. Excellent. Okay, let's
test this again. So now I can provide
general financial tips. That's perfectly fine because as you go about
prompt engineering, you'll notice that some prompts
on certain professions, artificial intelligence
cannot fulfill that role, so it's going to disclose that responsibility
like a disclaimer, but it will get the job done. So now we have our prompt ready, save the custom prompt now, we're going to be selecting different parameters
for our prompt. You can go quite detailed
with the process. I'm walking you
through it briefly, and it could vary
from one application to another or from
one case to another. So we give an action,
we create a prompt. Now let's select the parameters. So review actions parameters, review input and output
values and update descriptions as needed.
So what is the output? It's going to be a text, an image, whatever it's
going to be a text. Let's say, keep it as a text. Okay, next, review and test. Now, let's test our action. Now, make sure you test your action before creating
and publishing it. Verify it works
with your copilot to avoid errors that
could affect your users. So the name of the virtual
agent, financial advisor, does the description so now you have the
ability to test this, but it's going to be testing it directly into the available
option or channel, which is Microsoft Teams. So I will deploy it over there. We don't want to
deploy randomly. So we're going to click Publish because we
have tested this. So your prompt action
is being published. This might take a few seconds. Then we have the ability
to actually take the virtual agent and
deploy it anyway. Because once you run the test, it automatically
open Microsoft Teams and try to run it over
there to test it. But like I mentioned, the process is quite
straightforward, where you add
action details what you would like the butt to do. Create a prompt,
select the parameters, and review and test it,
and then you publish it. Your prompt action is now published to co pilot
for Microsoft 365. That means it has
been integrated into the various tools
within Microsoft 365, such as Microsoft Teams. Let's click on
GoTo Details page. It will navigate me to
copilot for Microsoft 365. So it means this
copilot now has been integrated as an assisting tool within COP for Microsoft 65365, you click on the
Export solution. You have the ability
to publish this. Now I'm going to click
on Publish here we go. Now, this will deploy it as part of Microsoft 365 package. So you have to
keep this in mind. It's not just simply
a custom pilot that you're going to be adding, it will be part of the
copilot for Microsoft 365. This is one of the
things that you have to keep in mind
when you are using artificial intelligence to
build your virtual agents. Once you go for the AI version of the development process, it will connect it
to your copilot for Microsoft 365 instead of a custom pilot where you just simply deploy it to any
other item that you'd like, whether on a website
or an application, it will integrate it within your tools within your co
pilot for Microsoft 365. So you have to keep these
things into account. It will keep it as a solution and you have the
ability to export this. So this is something to keep in mind if you're trying to build a virtual agent using the prompt action because once you deploy it
towards the end, it will not be treated
as a custom copilot, like the ones that
we have created, and then you will be
able to just simply deploy to any of the channels
directly and publish it. It will be deployed across
the Microsoft 365 directly. So now you have a general
idea how you can go about building your
virtual agents for various applications, whether through a template or building it from
scratch or using Artificial Intelligence to
develop the virtual agent and deploy it across your
Microsoft 365 suite.
12. Wrapping Up: So what do you think? I truly hope that you found
the class helpful if it helped you get up to
speed, get that knowledge, get that expertise with
fraction of the time in terms of learning how to
create virtual agents, assistants, AI chat bots, then it's a job well done. Often, you tend to find that these topics are quite complex, and the whole purpose of the
class is actually to make them easy for you to go from A to Z and go ahead and dive
into the application without the whole research and the whole complex learning curve that
you might come across, specifically when
you're learning about such important concepts which are new, which are innovative. I look forward to receiving your feedback on the
current class and make sure that you follow
my profile for the latest releases and updates, and I'll see you
in the next class.
Engr. Hussein Attié, Entrepreneur I Engineer I Educator
