Chapter 3: Conversation Design Practice
Learning Cognigy.AI in just minutes
This course uses Cognigy.AI as a Conversational AI Platform. Thanks to its low code
approach, it’s perfectly suited for Conversation Design. Furthermore, with all its
features and extensibility it’s also a great choice for creating fully-featured virtual
agents or phonebots. Anyways, if you are already familiar with it feel free to continue
with the next parts. I don’t wanna bore you and don’t mind, the course is also of value
for you if you or are using another platform or toolchain.
Conversation Design and Development go hand in hand. No matter if it’s for a text-
based channel or for voice. Cognigy.AI supports you with that through all phases from
creation to maintenance. Just think of a Content Management System for
Conversational AI. It takes just minutes to understand the basics – that’s one of the
reasons why I love to use it in workshops teaching Conversation Design. All you need
are Intents –understanding what the user wants to achieve, Reponses – the bot’s
answers or replies and, finally, the Publication – making your bot available to your
users on your preferred channel.
So, once you have signed up for the Cognigy AI free trial at https://signup.cognigy.ai/.
You’ll create a virtual agent in your agents' view by clicking on "Create Agent". In the
next step, you give your agent a name, I chose cai for conversational AI. I select my
favorite color and I keep the selected language English because I want to use English
for the natural language understanding. Now I click next. And he can choose a template
for my virtual agent. I stay with the blank template because I want to start from scratch.
And in the next step, I can select a channel. That means the way I like to publish my
project, my bot, my virtual agent to the world and I keep Webchat here. And the next
step I skip is because I don't want to add some additional capabilities like Smalltalk or
the detection of celebrities right now. And I click on Create, which sets up my virtual
agent. Once this is done, I click on Chat with your agent, because this leads me directly
into the chart view of my bot where I can see how inputs are processed and at the
same time, on the right, I see the chat window, which I can open and close by others
with a call-out button to try out my conversational AI, to try out my bot. Here, I type
something like “Hi there” and send it to Cognigy AI, to my bot, to my conversational AI
and get the response “Hello World”. Because he has only one Say Node which
response to everything, saying Hello World.
So how can I turn this into a real working bot with a lot of different answers? All I need
are intents. Intents are set up through the NLU tab here, the natural language
understanding tab, and here I can create my very first intent. I name it: What is
conversation design? Obviously a useful FAQ, a useful question to a bot on
conversational AI. I now can save this intent and add example sentences. Those are
the examples that make up the intent which help to train the model to understand
similar questions related to conversation design. So I write something like the phrase
conversation design. And maybe a question like what is conversation design?
I save those samples, and now I have to make sure that the model gets trained, so I
click on Build Model. This takes a few seconds. And I now go back to my chart view.
And in my chart view, I need a way to distinguish between several inputs, between
different inputs. So I click on this + here and I select Lookup.
Lookup makes exactly that. And it's set to intent. So that means it looks up the intent,
the detected intent. And it has a default branch for "I have no intent detected" and a
couple of cases where I can choose if I select those. Where I can choose the pre-
configured intents. I can save this node. And now I can add a different answer to this
node. If I click on the plus, I choose the basic Say Node. And for my basic Say Node,
if I click on it, I select an answer like: conversation design is great. What else do you
like to know? And I save this Node. And before I try out my bot, I make sure that I don't
get two answers in a row that are not related to each other, so I move the Hello World
Node to my default branch.
This means that if I know right something like blah, this goes into my default branch.
And if I ask for conversation design, this goes to my What is conversation design
branch? Because the intent is detected. And if you like to extend that with your own
ideas, simply add more intents to the NLU, train them and add more cases with Say
Nodes which have the response. This is the way you build a simple FAQ bot. And if
you want to see what it looks like in a real Webchat, go to the left, select the deploy
menu, click on endpoints and you see the Webchat. If you select Webchat, you can
change a couple of things. But for now, I only click on Open Webchat. And here you
go, you can again try out what you built. Blah leads me to Hello World, to the default
branch. And something like conversation and now let's try the AI we are using the
Natural Language Understanding. Instead of design, I write conversation art. And there
you go. It's detected even when not used in a sample sentence. It's close enough to
design so that we get a proper answer.
That’s all you need for now to start with your first Chat or Phonebot using Cognigy.AI.
And don't forget to check out the next parts on Conversation Design, to build even
better bots.
