john-john.nu is running on Pythons Flask application. Using a little bit of Tailwind and a whole lotta JavaScript
the idea was to give it a livestreaming vibe, only where the actual streaming is done by my avatar.
For the logic of the avatar I am sending API calls to a ChatGPT 4o-mini model that has been pre-prompted
with some basic information regarding my references and typing traits.
The ChatGPT response was thereafter ran through a text-to-speech API service hosted at ElevenLabs, where I once again
trained the model to resemble myself but this time with my voice instead of reasoning. As a sidenote I am very happy
with hearing myself speaking english with seemingly perfect pronounciation. [Edit: The speech functionality has been temporary disabled due to subscription costs.]
Once all that was done I just had to record a minute of myself typing away at night in front of the computer.
The demo video is not recorded by me, it is made with an AI avatar progeram following a speaking script I wrote, I was very impressed
with it seeing as it only required one minute of uploaded content to train the model. The company whose model
I used for this was HeyGen.
All in all I am using a minute of video and a minute of recorded audio for the total training of AI models.