Create your own Copilot AI engine using Copilot studio

Microsoft Copilot Studio is a great platform for creating and managing custom copilot. You can make great interactive models, without even needing to write a single line of code and allows users to create and deploy intelligent, intuitive, and connected AI assistants for employees, end users, and customers. These copilots could be tailor-made to suit all your business needs. This article is a short demo to create and deploy one such custom copilot using Copilot Studio.

So, here are the steps to implement the same:

Step a: Before you begin, you can have a demo instance ready by spinning it up from: Dynamics 365 Project Operations (CE) - Preview Trial - Sign up (microsoft.com) and fill up the wizard. It will create a ProjOps instance with D365FO connected to it. Allow approximately 1 hour to complete/prepare the instances. Once done, you can launch https://admin.powerplatform.microsoft.com/ to see the environments. But do remember to select: US as the region -- else a lot of things won't even work for Copilot. 

Step b: Logon to: https://copilotstudio.microsoft.com/ to start >> logon with the enterprise ID which you created from above step and voila! -- you will get this as the landing page:

This is where a lot of OOB templates are available for creating various Copilots. Additionally, we can create Copilots from scratch (click on the Create side button), there are innumerable libraries to choose from too (click on Library side button).

Most importantly, you can come to your home page and start typing your Copilot from the search text itself:

Here we are inputting our requirement to the studio in our conversational words (NLP). Press enter to continue.

Step c: The following screen will land up:

Here it's expecting you to fill in the kind of conversation you are expecting your Chatbot to initiate; on the right-hand pane, you can configure the language with which you want your copilot to interact. Essentially this how the LLM (large language Model) is interpreting the inputs from the user (RAG), to Ground into the resource availability.  

Evidently, the above chat is asking if you want to specify the tone of the chat. And if you are not sure of what to answer, click on 'What should I say' -- the chatbot will open up a series of new suggestions:

Which means, the ChatGPT algorithm running behind the hood is further breaking down the code into further rephrasing and finding more granular specific commands.

Following this, it's asking any 'method' to search for books, to which I responded with URLs like Amazon, Literature-map and GoodRead to read from:

And then I conclude the wizard chat by saying: 'I am done'. When I said that -- it automatically opened a side pane to test my chatbot:

Let us find 'Da Vinci Code', the famous novel by Dan Brown:

To which the reply is:

Which is a link from Goodread.

Walah!!! Now let us go and create this Copilot by clicking on 'Create':

Step d: This Will take a while to complete, and after some time, you would see the vibrant workspace has created the copilot like this:

Which is a totalitarian view of all the available capabilities of the copilot. I can move over to Overview and edit the changes for my Copilot (eg: renaming it):

I am giving a fancy name to my Copilot:

And also a photo icon for the same.

Step e: By default, I have given three websites to pick/search the data from. However, I can add more knowledge to my site to get 'Ground'ed with more data. Click on 'Add knowledge' button to continue:

There are a number of choices to add from -- though some are marked as 'Coming soon'

Let me add from Dataverse which can allow me to add from Dataverse tables, previously/already created. I have already created a dataverse table called Books, which I am going to select here:

Which would allow you to add by pressing Next to finish. Now our Copilot is ready with all the necessary knowledge:

Allow it sometime to pass necessary validations, to read it and register it as a valid connecter:

Additionally you can also add files to your Knowledge base (CSV, excel -- anything that contains stacks of information). You can also tell it to search from public URL, SharePoints and OneDrive units:

And not to mention -- a litter of connectors to consider to train your Copilot. In fact, literally, it can pick and consider any source to learn from. 

Our Copilot is ready, per see. However we can still enrich the experience by creating/adding actions to our Copilot experience. Watch out my next blog for more details.

Whew! So much of Copilot for today -- will return on more such cool features soon. Stay connected and Namaste.....