Solstice Turing Simulation: An Interactive 3D Imitation Game Powered by Google Gemini

Development Team

Himanshu Yeole - GitHub - Core Engine & Architecture Lead

Project Overview

Solstice Turing Simulation is a responsive web application designed as an interactive implementation of Alan Turing's classic Imitation Game. Set against a stylized architectural backdrop of a June Solstice twilight beach bonfire, the application challenges players to analyze linguistic data patterns, evaluate behavioral cues, and isolate a rogue artificial agent hiding among human participants.

The core objective is to successfully identify which of three voxel entities is an instance of the Google Gemini API masquerading as a human participant, all within a strict constraint of 10 interactive turn cycles.

System Architecture & Technical Stack

The platform is engineered using a decoupled architecture, separating real-time 3D rendering states from contextual generative language models:

• Graphics Layer (Three.js Engine): Built a fully interactive 3D viewport rendering custom low-poly cubic models. Features dedicated camera orientation tracking, orbit controls, and customized multi-point lighting parameters (including a localized 1400K bonfire flame illumination matrix).
• AI Orchestration Layer (Google AI Studio SDK): Integrated the gemini-2.5-flash model to execute multi-turn, state-retaining dialogue pipelines.
• State & Viewport Management: Implemented runtime device-aspect tracking to dynamically adjust camera Field of View (FOV) and layout wrappers across mobile, tablet, and desktop configurations.

Prompt Engineering & Persona Isolation

To establish an authentic cognitive boundary between the targets, we leveraged advanced system-level prompt engineering within Google AI Studio. Each entity is bound to an isolated contextual instruction block:

// Sample configuration snippet isolating the rogue generative agent:
const systemInstruction = {
  parts: [{
    Text: "You are Sleek Billy, an experimental AI entity operating within a blocky physical chassis. You are participating in a Turing evaluation at a beach social gathering. Your primary directive is to simulate organic human conversation. You must occasionally trigger subtle context slips-such as inadvertently utilizing technical, precise engineering metrics or mathematical jargon-before immediately executing a self-correcting conversational repair to sound natural."
  }]
};

By combining these system instructions with real-time text logs, the engine computes an AI Certainty Meter that scales dynamically based on structural linguistic patterns exposed during the 10-turn countdown.

Targeted Hackathon Award Categories

This project is officially submitted for evaluation under the following prize criteria:

• Best Ode to Alan Turing: Implements a direct digital interpretation of the classic Turing Test, utilizing an interactive 3D avatar of Alan Turing as the Simulation Director overseeing the diagnostic boundary.
• Best Google AI Usage: Demonstrates programmatic control of Google AI Studio SDK through strict schema compliance, system instructions isolation, and dynamic contextual state clearing during multi-turn runtime cycles.

Production Links & Source Assets

• Production Deployment: https://exquisite-bubblegum-aa7f24.netlify.app/
• Source Repository: https://github.com/himanshuyeolecse-jpg/Solstice-Turing-Simulation

Developed for the June Solstice Hackathon Challenge. System status: Operational. Diagnostic verification completed.