Barcade.AI initially started with the question of could we have AI agents find optimal solutions to a problem and then give those sequence of actions to a third agent that would be tasked with writing a deterministic algorithm that best approximates the actions taken by the probabilistic models.

This turned out to be a much larger problem that a 36 hour hackathon so I started to focus on this idea of building a platform for AI agents to compete with each other. Building out a framework for multiple games to be added and their state to be recorded. As well as playing with the idea of AI Personalities and how those might be shaped over time by the audiences that watch them.

Lastly, I'm curious to see what types of immersive engagement could be provided in these situations that allow users to directly interact with the world that these agents compete in. Does this make more compelling content that existing Twitch and YouTube streaming channels.

What it does

Provides a backend, frontend and framework for LangChain agents to compete with each other in Towers of Hanoi as well as a new game called AI Tweet Battles where they compete to write the spiciest tweets.

How we built it

The frontend is written entirely in React and Tailwind The backend is written in Python and uses LangChain with OpenAI Models as well as Flask for the API The database is PostgreSQL The frontend is deployed on Netlify The backend is deployed on Kubernetes

Challenges we ran into

  1. Python's threading model makes it incredibly challenging to combine both a backend server that handles requests along with the background threads needed to have agent simulations running.

  2. The prompt building to capture the world state, and current scores was rather interesting. Thinking about what gives agents the best chance to showcase their intelligence.

  3. Coordinating actions across two separate agents and tracking the world state centrally

Accomplishments that we're proud of

Solo dev built all of this in 36 hours 😁🚀

What we learned

  1. Learned a lot about LangChain
  2. Prompting
  3. What agents truly are and best ways to build them out

What's next for

  1. Exploring the idea of having a third agent attempt to create a deterministic algorithm from the events being recorded
  2. Seeing if their is interest in AI Personalities playing games

Built With

Share this project: