This project comes alongside an article that I am still currently writing, here is its introduction:
A few weeks ago, I started playing on a Minecraft server. To have a new player-experience, I searched for new mods and I came across the Neighborly mod. Neighborly is a really nice mod where you have some "friends" in your world. they all have a uniq personality and we can interact with them by giving them gifts or building them a house. I had a lot of fun playing with this mod but after a while, I found all of them and built a house for each one and the fun kinda stopped. That's when I had an idea.
Everyone speaks about LLMs and artificial intelligence in general. They offer endless possibilities and the ecosystem around them was growing. What about using them to manipulate the entities of the Neighborly mod ? so the big question was: Can IA be creative enough to interpret a character with a uniq personality and how ?
I used this repo to test stuffs. Lately I used it to play with Langchain.
I will migrate to a monorepo soon
The software architecture is bad, I was testing new things and kinda got lost. It surely can be improved.
- Create a new companion with a background
- Have a conversation with a companion that will have a personality based on the background
- History is persisted
- A summary is created from the history to avoid sending the history for each request
- Integrate all repos in a monorepo (probably maven since it's mostly Java but ideally this will be used by a Minecraft mod so also probably gradle and I really like NX so probably NX)
- Implement the CRUD on the web-client side.
- Add auth (probably with Auth0)
- Deploy (Idk where yet, but I'll probably just provide the docker images and a docker-compose file)
- Finishing the article on the subject
- A lot more, I have a lot of ideas on this subject