It's inevitable with today's technological pace that our digital footprint will continue converging with our physical selves - I thought I'd jump the gun and just publicize a digital version of myself.
Simple Setup
The implementation is quite straightforward. I post-trained a model using a few .txt files that I filled with my personal backstory. The information is embedded locally using All MiniLM-L6-v2, and the chatbot then runs on the GPT-4o model. My server is already strapped for capacity to run a fully local model properly—perhaps a project for another weekend.
You can test it out by clicking on the chatbot icon in the bottom right. It's rate-limited, and I haven't trained it on any sensitive info, but feel free to battle test it ;)
Future Potential
In this rudimentary form, the chatbot is pretty bland and basically only useful for entertainment purposes, but it serves as a nice proof of concept. The threshold to plug into strong models with personalized content or api access into more applications will continue to become more frictionless. This gives me strong conviction that the most useful infrastructural applications will be the ones that happen as close to the OS as possible (though this also adds security and privacy concerns that need to be considered). The new release of Raycast AI materialized this for me as a clear progress in this direction - what might seem like another gimmicky rebrand is genuinely changing my workflow.