Can you solve the mystery of Albert's death? SeagullStory is a text-based detective game built using Pygame, powered by advanced Natural Language Processing (NLP) models. The game is based on the famous lateral thinking puzzle: a man orders seagull soup, takes one bite, and immediately ends his life. Players can type natural language questions to uncover the truth.
It is powered by a custom dual-LLM system to understand and answer any boolean (Yes/No) question you ask, as well as to guide you through the story with hints and unlocking scenarios.
This project uses a hybrid Cloud API + Edge AI architecture:
- Primary Model (Cloud API): Powered by a DeBERTa-v3 model fine-tuned on a custom dataset relative to this story and hosted on Hugging Face Spaces. It accurately answers user questions with "Yes", "No", or "Doesn't matter".
- Progress Tracker (Local Edge AI): Powered by a highly compressed, quantized xtremedistil model converted to TensorFlow Lite (TFLite). Weighing only 12.4MB, this local model tracks player progress in real-time and allows player hints.
- The NLI Trick: Progression tracking is framed as a Natural Language Inference (NLI) task. The system checks if the user's question entails hidden key facts, dynamically unlocking new story scenarios as you get closer the truth.
💡 For more technical details
Read the full Technical Architecture & Model Training procedure.
You are given the cryptic final paragraph of a story. To win, you must reconstruct the events leading up to Albert's death by asking natural language questions.
The AI will respond with:
- "Yes" — if your hypothesis is true in the scenario.
- "No" — if it is false.
- "Doesn't matter" — if your question is irrelevant to the story.
The story is broken down into 3 distinct, unlockable scenarios. Discovering key facts in one scenario will allow you to progress and grant you access to the next scenario.
Make sure you have Python installed, along with TensorFlow (required for the local TFLite progress tracker) and Pygame.
Clone the repository and run Game.py






