This is the backend part of https://ai.repl.page.
The OpenAI model being used is text-davinci-003, trained with:
- Major parts of the Replit docs
- The Replit blog
- The Replit landing page
- The Replit Employee Organization Chart
- Amjad's personal blog
- Amjad's AmA Repl
- Amjad's Podcasts
- Some of Amjad's Tweets
- Replit's Terms of Service and Community guidelines
- Some random facts the AI should be aware of
All of the training data should go in the facts folder, and then run python process_data.py to populate the .index file.
For more information on how this works, check out Zahid Khawaja's Tutorial.