Breath-Mint

• 

  The standard interface for the game.

Vision

We see people care deeply about how they act and what they say when playing videogames, and that same pattern appears when trying to get an LLM to understand exactly what you want from it. With our experience in AI development, we decided to constrain models on complex player input and game state data. At that intersection, to fine-tune that pattern to tease out one's dearest emotions, we are building our product(the models, tools and Chain), in the form of text and art and mechanics and heart.

So, Why a videogame? Also, is it fun?

We are building the modern iteration of a choose-your-own-adventure, but with a LLMs working in tandem to tell a wholly human story, of the upcoming Korean population collapse. With an active backend structure that removes hallucinations through data, tool use and a multi-agent gameplay system, the vision to build our game also describes the foundational structure to precisely recall and update dynamic information. This structure is extended into the saves of the game, so players combine their saves to create playthroughs, minting them, sharing and spreading them, the value coming from their own words, through the state-machine of the game.

We are also building no-code tools for all artists to tell new kinds of complex and interesting stories, further expanding how many people would write and read these new kinds of stories. The data generated from these games, especially from the multiplayer structure, would be valuable, novel, important. Unlike telling an LLM to simply try again, this would be decision making data, time-bound, data that helps us understand the very core of videogames and entertainment (which means, extremely valuable). So, its not only fun, but valuable!

Who we are

We're a fairly small but very driven team, each with our own expertise. Vaibhav Girish is our AI expert, with experience building tools for developers and applications interacting with the bare-metal layer of computing. Shubhang Dwivedi is working on a high-velocity backend service, drawing from his experience at Urban Company. We're designing our frontend and database structures from the best practices available, using vectors for our databse, done by Alister Rodriguez and Dhruv Umesh. Aatish Sahai is the person with the vision for the game, the story, the art and business models. He is also the only one applying to this hackathon because he found out about it at the 11th hour.

Challenges we're expecting

Gaming is in a bit of a lull, and crypto-games and AI-games are feeling the slowdown more than anyone. Crypto MMORPGs and play-to-earn games have ruined the perception of the ecosystem of Crypto-Games, for most people outside the crypto space. AI corps themselves are running out of data, with 50% of ChatGPT's responses coming from Reddit and Wikipedia (neither of which is very good), and yet have a stranglehold on the market share, with others left to partnerships and technical boundaries (neither of which are ideal for LLMs). The intensity of investments into AI to solve coding once and for all isn't bringing the expected returns either, mostly due to the mathematical constraint of hallucinations.

Accomplishments

The LLMs we work with are great at text generation and parsing, at context-specific execution of code, and keeping the "flow" of stories going. We're proud of the stories we've written and the format we've created, as well as the art, pacing, drama and overall narrative of our bonkers sci-fi story.

What's next for Breath-Mint

Raise investment to scale this project, so many humans making many videogames using many AIs have many fun.

Built With

• claude-api
• flask
• python
• railway
• redis