Splendor AI

I've been losing too much on this game. So I came up with the idea of training AI to train me to get better at Splendor.

Tech Stack

• VITE React on TypeScript for frontend.
• FastAPI for communicating with the backend.
• Homebrew my own ORM and auth (?) if I'd like to deploy this for HvH or HvM.
• Use Gymnasium API to set up the game environment.
• CNN would be the initial model for the AI. Would venture to other models if it goes well.

Roadmap

• import Splendor cards and nobles data
  • cross-check cards with visuals
• compose game structure and rules
  • present initial game states
  • validate some game states
• compose game API (done-ish)
• make AI model
• train
• also do some unit tests 'cause every good boy does that

buildout

Very first thing to do before starting:

• please run ./run-me-first to install stuff for first time use.

on backend/

• run using ./serve

on frontend-ministry/

• run using npm run dev
• you can also port forward, but with this config: npm run dev -- --port=5174 --mode=tunnel

cards_minified.json

source https://www.scribd.com/document/534963546/Splendor-Card-List-With-Pics

conversion

cards_minified.json >> python scripts/generate_cards_json.py > cards.json

helper app

websocat

• https://github.com/vi/websocat/releases
• sniff data (example):
  • ./websocat.exe ws://localhost:5000/ws/train