Microsoft 365 Copilot design and architecture

Brass Contributor

While there is still not too much public information I decided to create a simple design of the different Copilot features together. Much of the Copilot mechanisms is dependent on the Microsoft Search feature in Microsoft 365 which will build a semantic index based upon data that is being indexed from Microsoft 365, but can also include 3. party data connected via Graph. You also have custom federated data sources into the Search Engine from data sources such as Dynamics or Azure Cognitive Search. 

MariusSandbu_0-1687987331646.png

 

Will create a new post with the visio once I have all the pieces in here! hope this helps to give you some overview of the feature.

 

8 Replies
This is awesome! One thought....the MS Graph needs representation here. Not sure where though
Thanks! will add it, have made some adjustments based upon new information in regards to the indexing and how the LLM traffic is being processed in the different regions.
I am struggling to find how Data is processing in data at rest and transmit in copilot architect? Do you have any idea ?

I am struggling to find how Data is processing in data at rest and transmit in copilot architect? Do you have any idea ?

 

@Laxman V Sure let me explain. If you are using CoPilot as part of M365, and we use one example where we are just using CoPilot to chat with our data.

1: The end-user sends a prompt "How much vacation am I entitled during summer?" into Copilot
2: CoPilot will then alter the prompt with its own system message, not sure of the specifics here, but that one we will know more about later. 

2: The Copilot service will then trigger a search using the Microsoft Search service, which will then in turn use the Semantic Search Index to try and find data that is similar to the initial prompt. The search here is similar to the capabilities that are present today in Microsoft 365 with regular search. All search is done in the context of the access that the user has. 

3: The service will then present results with the data that might provide answers to the initial prompt
4: The results and data are then sent trough the Azure OpenAI service for Copilot which then generates a response back to the user. "you are entitled to X amount of days, based upon these sources" 

So much of the capabilities is powered by Microsoft Search, which gets information from the prompt and the results are then processed trough the OpenAI LLM. So data that is presented to the user is only data that the user has access too. 

Does Copilot index content from my OneDrive or Sharepoint sites I have access to by default? I would think not as that will be too much data. Any reference to the details on what Copilot indexes by default?
Hi Marius, this is very helpful. Were you able to create a Visio and could share here for us? Thank you.

@Laxman V Thank you. Will look into this.