Catana

Vibe coding
Fully vibe coding
Analyzing board state
Marven lost against AI
Our sample2 board that we analyzed in the demo

Inspiration

Our friend Marven plays Catan 40 hours a day. He got so clinically good that he got tired of beating us in person. So he decided to build some minions for some asynchronous beating.

What it does

The minion uses multi-threaded AI agentic system with async call management to swiftly deliver the asynchronous defeat.

How we built it

We started with data preprocessing which embeds statistical meaning to the board state. Then we used multi-threaded AI agent that we designed ourselves to mimic top player logic and decision process (aka Marven) with extended simulation space.

Challenges we ran into

Our prototype AI structure took too long to respond to our query -> We optimized call orders and reduced dependency between agents by segmenting data and tasks, reducing total synchronous layers per decision.

Accomplishments that we're proud of

Our final model was able to generate the optimal solution to our case studies for all 6 of our sample board states within a minute while we took over 10 minutes to derive each optimal solution.