Raspberrypi DM. DUNGEON PIE

• 

  actual hardware it self

• 

Inspiration

One of time i had the most fun in life was playing dnd with my friend and When the dm did not made it to the session i was devastated thus born this idea.

What it does

Right now it is a prove of concept providing a simple text base adventure to beat the dragon. with the most generic story line. and you have to find a way around the minotaur to kill the dragon

How we built it

Using NER and sentiment analysis to perform NLU and intent recognition. the code will try to respond to the user English input as best as it can with limited compute power.

Challenges we ran into

Compute, RAM and Time Complexity of natural language processing algorithm. below is a non exhaustive list of the problem i had to overcome or find work around

1. Finding the model that can actually run in a reasonable ammount of time.
2. Finding the model that can actually fit in to the raspberrypi limtited ram
3. Realising that LLAMA CPP vulkan backend dose not work properly on rpi5
4. tried to train MAMBA SSM model to perform NLU but the size of the lora and the parameter was too small
5. MY LIMITED ATTENTION SPAND ## Accomplishments that we're proud of I had fun and i got an idea. This idea was partially implemented enough to see if it works or not. ## What we learned MAN AI IS EXPENSIVE TO RUN ## What's next for Raspberrypi DM. DUNGEON PIE To be honest i am prob better off sticking with a more powerfull general purpose llm with vision support since i have so much more idea to implement but can't due to lack of compute resource or time to wait for the code to run

futher more i wanted to add a way to implant real world object into dnd world and wise versa thus the usb webcam

Built With

• huggingface
• llamacpp
• nltk
• ollama
• spacy
• trl