Inspiration
The hardest part of assembling a DND party is often finding a Dungeon Master to lead the game. In the world of AI, we can give this task to a DM we can tweak to our liking and generate engaging scenarios.
What it does
We have 3 parts to the system. We have a simple web frontend where players can type their messages to the DM. We have the backend which holds player data and scenario history. And finally we have the multiple AI agents that interact with each other and to create responses based on player actions. Each agent has a specific role such as combat, lore, environment.
How we built it
Frontend - React, Next.js
Backend - Python, PostgreSQL, Flask
APIs - Google Gemini
We used VScode live share to work in collabration. And GitHub for version control.
Challenges we ran into
We had some trouble passing data between the frontend and the backend while also running the AI agent. Sometimes the AI agent can want really specific prompts and get stuck.
Accomplishments that we're proud of
We can have multiple player characters interacting with the AI agent. And the agent can engage with npcs/environment and the players in a realistic manner. For a lot of us we are using quite a few new technologies. We have been able to integrate all parts together. We added AI image generation from user messages.
What we learned
How to train multi-agent AIs Connecting frontend and backend, including learning the tools to build each Using SQL within python
What's next for DND Companion
If we had more time: Add AI voices for each response. Refine prompts and responses from the AI. Clean up code by using best practices.
Log in or sign up for Devpost to join the conversation.