LLM Controlled Board Games
Explore how large language models can transform tabletop gaming. AI takes on the role of a dungeon master in Dungeons & Dragons and judges humorous card combinations in Cards Against Humanity.
Use Case
AI-assisted tabletop gaming, automated dungeon mastering, and AI-based judging in card games. The project demonstrates how language models can support game facilitation, voice interaction, and interactive entertainment for households, schools, and gaming communities.
Target Audience
Board game enthusiasts, Dungeons & Dragons players, households, schools, and groups of friends interested in AI-assisted gaming.
How It Works
The Dungeons & Dragons implementation uses a modified version of the open-source PlanarAlly platform connected to a Python backend. A virtual player interacts with the PlanarAlly web server to perform tasks normally handled by a dungeon master. Spoken commands are processed using Whisper and OpenWakeWord, while Kokoro generates spoken responses.
In Cards Against Humanity, OpenCV and PyTesseract are used to recognize text on playing cards. A large language model evaluates the submitted cards using additional instructions, selects the funniest combination, and explains its reasoning. Kokoro provides audio output.
User Journey
- Launch the desired game.
- Initialize cluster access and the backend system.
- Project the game onto a tabletop.
- Play the game with AI-assisted interaction.
- End the session and restart the game when desired.
Technologies
React, Python, Raspberry Pi, Qwen 3.5 34B, Cluster API, PlanarAlly, Whisper Large v3, OpenWakeWord, Kokoro 82M, OpenCV, PyTesseract
Author
Tuuli Silvennoinen, Sampo Perunka
Project
GPT-Lab Summer Internship 2026