Inviting Large Language Models to be part of our conversations.
Note: The name ScottAI is only temporary, it may/will change at any time
This project is currently finishing up a major refactor, apologies for the spaghetti code that is yet to be cleaned up!
- A bot that is invited into a conversation (in this demo, a Teams call)
- Given a virtual voice and virtual webcam so it can interact with others
- Access to multiple knowledge sources & skills, and can be extended for different use cases
- Voice, Video, and Chat access, allowing for multiple channels to be leveraged at once
Note: The entry point in this repo is in the ScottAIPrototype.Console folder
- Clone or download this repo
- Deploy the Bicep or ARM template located in the deploy folder. Eg:
az deployment group create -g myResourceGroup -f deploy/deploy.bicep - Update the ScottAIPrototype.Console appsettings.json using the values from the resources deployed.
- Update the meeting.txt file with a link to a teams meeting. This file should contain the URL and nothing else.
- Start ScottAI using Visual Studio, VSCode, or the following command within the ScottAIPrototype.Console folder:
dotnet run --runtime win10-x64 --no-self-contained -c Release - ScottAI should join the call once started, if you have the meeting not set to auto-accept guests it will prompt to accept once in the lobby.
To add
- What�s up with the virtual webcam image? This started as a technical demo, and while this is currently a placeholder, it is designed to demonstrate how personality can be shown through real-time rendering.
- How does the skill engine work? / Can it do X? There is a very simple low-latency skill trigger system within the code that can be extended to interact with any data, API, or anything you can code. It is also designed so more complex frameworks like Semantic Kernel can be swapped in.
- I can't get it to run/it crashes! Check the getting started guide, there is a specific runtime required for the native dependencies.