Event details
Is there some high-level guidance on which Microsoft AI solution/tool to use, under which circumstances/requirements, for building a RAG based Chatbot?
I found a few different options in the Microsoft documentation and understand some are low & no code options, but what is possible to do with them & what is not, is hard to understand from the documentation.
- Co-pilot Studio (PVA) based (No-code)
- Azure OpenAI Studio --> Chat completion --> Deploy default App (Low-code)
- Azure OpenAI Studio --> Chat completion --> Deploy to PVA (No/Low-code)
- Azure AI Studio --> Build your own Copilot
- Azure Solution Accelerator
- From scratch using Python / .NET / etc.
MicrosoftHi Vivek It really depends on how much control and customization you want. The specific technology powering the platforms is generally the same.
The intent for Azure AI Studio is that it will have parity features with Azure OpenAI Studio, and at some point, replace Azure OpenAI studio.
The intent is for Copilot Studio to replace Azure OpenAI Studio --> chat completion --> deploy to PVA.
I believe feature/capability is already at parity, Copilot Studio may have more available capabilities now.
From a RAG perspective, Azure AI Studio, Azure OpenAI Studio and Copilot Studio all use the same feature: Azure OpenAI on your data. This makes it easy to quickly ground your app with your own data.
If you require multitenancy, you should go with AI Studio. If you're building in a single tenant environment, you should go with Copilot Studio.
If you are building a custom copilot for internal users (e.g. an human resources knowledge base), business operations or creating a customer engagement solution, you should probably start with Copilot Studio.
If your chatbot will be using significant M365 data sources like Sharepoint and Onedrive, you should start with Copilot Studio. If you will be connecting to OneLake, blob storage or other databases, you should start with AI Studio.
You technically can connect any data source for both Copilot and AI Studio, but one studio has an easier experience for certain data.
Copilot studio, as you mentioned is a low code based platform, and provide out of the box chatbot interfaces/apps to use.
Azure AI Studio is in preview, as is the SDK, so you may prefer starting with a solution accelerator if you prefer to code.
Hopefully this helps - let me know what tech specifically you are interested in, and I can respond
- Thank you very much Allison! It is really helpful. :) At this point in time, we are in an exploratory mode and trying out a couple of Use-cases (interestingly HR knowledge base chatbot is actually one of them), internally focused, where the data is predominantly in our Sharepoint online and multi-tenancy is not a requirement yet. So, Copilot Studio seems to be a good starting point. As a EU based company, we have some strict requirements for data residency requirements and need the Solution Stack (App, Model, Datastore, etc.) to be in the EU. As in the Copilot Studio option, most of the Stack is pre-built, I will take a quick look at the MSFT Documentation to understand if/how I can choose the Model / Model location and if I have any remaining specific questions I will trouble you in this thread again 🙂 Thanks again!