Open comments for this post
So I first started by scanning all of the cards so when the full app is eventually done it will have data to work with. I know most people would usually make the app and then get the data, and I agree that it is more logical but I wanted to challenge myself with this first.
Initially I tried using the Gemini API to scan the cards of the physical deck and put the data into a file/database, which worked fine and gave good results but I struggled with the rate limits it offers. I also tried using both Qwen and Gemma as a local LLMs to avoid said rate limits but they gave not so good results while also being quite slow on their outputs at least using LM Studio, and for 400 cards speed is really important.
I tried promting in both English and Hebrew but both gave non perfect results and were equally as slow (the video is at 4x speed):