Notes for Copilot Studio and Copilot
Notes for Copilot Studio and Copilot
Two parts are typically present in natural language processing.
First is intent which is what action do you want to do. Secondly it
is entities that define what the intent is acting on.
Within copilot studio, you can define the intent and entities when
creating your own copilot. You can define how it answers and
also specify specific responses for specific questions. From there
you can create a flow and link that with another application that
can handle the actions.
You may need to use a 365 copilot for things like finding specific
emails, answering general questions and having a wide scope
but still have the integrations with m365 apps and services. We
can use the custom copilots for these specific reasons:
Search copilotsimplementationguide by Microsoft to get help on
how to work with copilotstudio.
Copilot studio offers the capabilities to just simply give
commands to your copilot as context and it will create it by
managing the backend and infrastructure. This makes for a very
fast deployment of a copilot to be integrated in a companies
ecosystem. It can be part of Microsoft teams too as an addiontal
app to be used as a personalized company companion.
Under the actions feature, we can add APIs that allow the copilot
to take actions. This makes process automation duable.
When using a search engine to be used as a source for your
copilot, do not enable knowledge since it will make the search
engine useless.
The way the copilot retrieves data from websites to be used as
content is through indexing a simple HTML structure. Problem
with this is that modern websites typically use javascript to make
the websites more appealing and thus it makes it hard for the
copilot to index the information.
Data than can be fed into a copilot is knowledge (things like
excel, documents, sharepoint), topics where you define what
things to discuss (tailor convos), and actions where I can do
things with APIS.
When doing topics in copilot studio, we can add a question node,
which defines what it is you want the copilot to respond with for
a specific question.
When you use part of the new entities synonym as a question to
the copilot, it might skip the question and not register the entity
or topic you had.
the question isn't asked and is skipped. This is because you've
used entities and slot filling to retrieve the information from
utterance the user submitted. This approach avoids you needing
to ask the user a question that they've already provided
information for
Different types of variables in Microsoft Copilot Studio include:
System – These variables are normally populated with system
data. System variables aren't user-made and are part of the
platform.
Topic – These variables are user-made from either topic Inputs,
the Set a Variable Value node, the Question node or as the
output of other nodes or actions (e.g., cloud flows, HTTP
requests, connectors, custom prompts, plugin actions, etc.).
Global – These variables are user-made and are available from
any topic, and they're a good way to store data that multiple
topics use to help the conversation, regardless of how many
topics are triggered within it.
Variables are the best way to store dynamic data or data that
you want to perform conditions or checks on to drive
conversational behavior in a particular way, as you'll observe in
the next task.
Conditions directly work with variables which you can pick out
from your topics and responses of the copilot.
One thing with copilot is that when you enter one flow, its hard to
go back to another flow, it doesn’t have the ability to go back.
Maybe use an adaptive card a json file for better looking personal details.
Without a generative method, if the agent detects one entity, it will stay
there in that flow.
When tryung to browse and use a sharepoint file, it only detects the pdf and
not the excel file.
Indexing a scanned pdf might not be possible to do this. We can maybe use
power automate flow, make it into text and then use that as a word doc or
excel to be used in the knowledge.
This is also connected with a sales CRM to keep a history of the people.
Maybe use a copilot studio for sales to intergatr with our main copilot.
Amal sabri.
Take the users input as an email and send it to the email given (optional
Flow)
First one: make a flow for the user to ask a specific query and the chatbot
will reply back with the details of how to solve such issue. This info can be
taken from the company website itself.
Second flow: you can ask the user again for the details of the queries
themselves and then use that information to make the person get sent to a
specific number tailored to that query.
Do I include another subcondition even for the subcondition to talk about the
error type as its own thing?
Copilot ideas:
Americano notes:
To open a ticket, check the teams name if they have accces to it.
3 main options pop up in the selection, when selecting IT, you can select a
ticket number and add the number, from there add the service for it. To
resolve unlocking account issues, a link shows up to select it. You can make a
ticket if the issue still remains
You can ask service request and different options show such as sotware,
security hardware etc
Maybe have a decision tree early on in the flow to check if it’s a user or non
user.
To do:
Make sure you can build upon the project and add more use cases.