AI Battler

Inspiration

Many people think of AI and LLMs as something that replaces people, but LLMs are great at assisting people too. Generally, low-complexity and repetitive games get boring very quickly compared to higher-complexity games. We think that the solution to this issue is Generative AI to add replayability to a repetitive game loop.

What it does

Our game uses a script to ask Cohere's model to generate a list of enemies using predefined parameters. In this fashion, every playthrough will be generally similar but contain key differences which differenciate the way it is played. However, this behaviour may sound all too similar to a random number generator. The difference is that LLMs are context-aware, for example, they know that goblins are a small and nimble species and thus may give them the swiftness train rather than something else. On a higher level, this leads to theme consistency and thus a more immersive experience while also having high variability in enemy design.

How we built it

We built our game using Unity as the main game engine, running scripts of C# (which were also used to access Cohere API).

Challenges we ran into

One challenge we ran into was difficulty when it came to making API calls from Unity and also being able to parse the returned information into the format that we needed. We solved this by making use of resources like mentors and documentation.