Honky Harnold

We created a wildly chaotic, infinitely replayable AI text adventure rewarding you for making unhinged, animalistic choices and punishing you for using polite human logic.

Comment 1

Inspiration

We wanted to push beyond standard "choose your own adventure" text wrappers. Our goal was to leverage the reasoning and story-generation capabilities of modern LLMs to build a fully dynamic, infinitely replayable RPG with a twist: a deeply unserious, hyperactive goose Game Master.

What it does

Our game is a dynamic interactive story where your survival depends entirely on your tone. Instead of simply picking a logically "correct" action, the player must consistently choose the most chaotic, reckless, and "goose-like" options. If the player tries to use polite, sensible human logic, the AI punishes them and deducts health.

How we built it

  • Backend: We built a strict, referee-style backend using Python and FastAPI to manage the game state.
  • AI Integration: We used the Google Gemini API combined with Pydantic models to enforce strict Structured Outputs, ensuring the AI consistently returns valid data.
  • Frontend: We used TypeScript and CSS to build a responsive, animated user interface that communicates seamlessly with our backend API.

Challenges we ran into

Because LLMs are inherently stateless and creative, maintaining rigid game rules proved to be difficult. Initially, we struggled with the AI hallucinating game states, forgetting health points, or returning broken text formats that crashed our app. We overcame this by implementing extensive prompt engineering, utilizing a rolling history window for memory, and employing exact JSON schemas to ensure the AI aligned with our backend state.

Accomplishments that we're proud of

-Generating truly unlimited branching paths while maintaining a stable, unbroken game loop.

  • Creating an interactive, tone-based evaluation system that instantly reacts to the semantic mood of the player's choices.
  • Designing custom animations and backgrounds that bring the AI's dynamically generated fantasy environments to life.

What we learned

  • How to successfully bridge the gap between open-ended AI generation and strict backend data validation.
  • The intricacies of client-server architecture, specifically handling REST API requests and state management. ## What's next for Goose Game We plan to expand the game into a gamified educational tool for English language learners! By making the "winning tone" customizable, we can use the game to teach sentiment and nuance. For example, a teacher could change the settings so the player must identify and choose the most "calm," "professional," or "persuasive" options to progress, completely flipping the core mechanic.
Share this project:

Updates