"# AI-Sim

A browser-based 2D world simulation demonstrating client-side AI inference and training using TensorFlow.js. Published as a GitHub Page.

What it does

Two species of creatures live in a tile-based world rendered on Canvas 2D:

• 🌿 Herbivores — seek food, flock with each other, flee predators
• 🦊 Predators — hunt herbivores, hold territory

Each species shares a CNN brain (TensorFlow.js) trained in-browser via Deep Q-Network (DQN). Creatures perceive a 7×7 tile grid around them and learn which actions (move up/down/left/right/stay) maximise their rewards.

Player controls

Use the sidebar toolbar to:

• 🌱 Paint food sources
• 🟩🟦🟥 Paint terrain (grass / water / danger zones)
• Spawn or clear creature populations
• Tune the replay buffer size (1 = online learning, 2000 = stable mini-batch RL)
• Adjust simulation speed and exploration rate (epsilon)

Tech stack

• TensorFlow.js — in-browser ML
• Canvas 2D — rendering
• Vite — build & dev server
• GitHub Actions — CI/CD to GitHub Pages

Development

npm install
npm run dev       # local dev server
npm run build     # production build → dist/

Push to main to deploy automatically to GitHub Pages."