The code for the hackathon project is available on GitHub

I participated practically solo in the CoHere AI Hackathon to use an early version of the Cohere LLM to make an app. The project came in third out of thousands of participants and dozens of valid submissions. The project used real-time live voice transcription that was entirely browser based, HTML Canvas compositing for live webcam video input, and a fully custom RAG (retrieval augmented generation) code written in C# before RAG was even a known terminology. Information was retrieved from Wikipedia to help ground the AI’s responses.

I setup hosting for the app in AWS using the Lambda serverless architecture, which for C# is able to serve both the UI and process backend commands. The app was a fully functional prototype that could be used to record a screenshare and see AI responses in real-time.