README.md

metadata

title: Team 7
emoji: 🏢
colorFrom: purple
colorTo: yellow
sdk: static
pinned: false
license: other
short_description: final submission

MistralAI_GameJam2025

This repo hosts Team'7 project for the Mistral AI Game Jam 2025. Thanks to Mistral, Huggingface, Scaleway and ElevenLabs for providing the ressources!

Game Description

TODO

Architecture

How to Build the Front-End

This project contains three main components, the Python LLM middleware (using Django), Web interface (using Vite React), and Unity 6 game project.

To build the front-end code, you must follow the following steps:

1. Build the game

Open the ./unity folder through Unity Hub with Unity 6. Then, switch the build target to Web.

2. Prepare the build folder

When building the game, you'll be asked to choose the name and path of the output folder. Make sure you set that as gamejamproj, or all the contents of the Build folder has that name.

Next, rename the Build folder to build (only for consistency. You can change the build path and name in ./app/src/config).

3. Build the Web UI

Move the build folder to the .app/public folder. Go to the ./app folder, and run yarn to install the dependencies, and yarn build to build everything. The build output should be stored in the ./app/dist folder.

4. Move the Unity build to the web UI

Make sure the build folder that contains the Unity WebGL binaries are in the same path as the index.html file in the dist folder.

5. Run the page

You can deploy the page or run the index.html through a local HTML server. Note that due to WASM policy, you cannot just double-click the .html file and load it in your browser to play the game.

Backend

Personality-Based Decision Workflow

This flowchart illustrates the decision-making process for an AI personality-based word-guessing game. It visually represents how the AI determines its behavior and output based on its assigned personality trait.

Workflow Overview

1. Start Node: The process begins by parsing the BASE_PROMPT and inserting relevant context.
2. Personality Decision: The workflow checks the assigned personality trait (sensitive_to_compliments, rebellious, stubborn, lazy, normal, or overthinker), which determines how the AI interprets advice and makes guesses.
   • Sensitive to Compliments: Trusts advice only when it includes compliments.
   • Rebellious: Challenges or twists user advice, often defying direct suggestions.
   • Stubborn: Prefers to stick with previous guesses unless provided overwhelming evidence.
   • Lazy: Takes minimal effort, often choosing random or obvious guesses.
   • Normal: Processes advice straightforwardly and logically.
   • Overthinker: Analyzes hints from multiple angles, often second-guessing decisions.
3. Making Guesses: Based on the selected personality, rules and context are applied to propose guesses.
4. Satisfaction Rating: The AI evaluates the advice, assigning a satisfaction rating (0, 1, or 2) based on its helpfulness or relevance.
5. Output Assembly: A JSON object is created, containing the guesses, satisfaction rating, and reasoning.
6. End Node: The process completes with the assembled output.

Visual Key

• Purple (Decision Nodes): Indicate points where a decision is made (e.g., personality check, satisfaction rating).
• Pink (Personality Nodes): Represent personality-specific rules and behavior.
• Blue (Process Nodes): Represent operational steps, such as parsing prompts and proposing guesses.
• Green (Start) and Red (End): Highlight the start and end of the workflow.

This structured workflow ensures that each personality behaves uniquely, adding diversity to AI interactions. The color-coded nodes make the chart easy to follow and visually intuitive.