About OthelloAI Dojo

The trained AI opponent in this application is based on the seminal deep reinforcement learning architecture proposed by David et al. (DeepMind AlphaZero). It consists of an 8-residual block Convolutional Neural Network (ResNet-8 CNN V3) with dual evaluation heads:

• Policy Head: Outputs logit probabilities across all 64 board cells to select high-reward moves.
• Value Head: Evaluates position strength and estimates win probabilities from any given board state.

Model weights are stored in ONNX format (othello_model_final.onnx) and executed client-side via onnxruntime-web directly in your browser using WebAssembly.

This entire application was built with the help of an AI coding assistant. It's a demonstration of how AI can accelerate and enhance the development process. The following technologies were used:

• Framework: Next.js (React)
• Styling: Tailwind CSS and ShadCN UI for pre-built, accessible components.
• Generative AI: Google's Gemini models accessed via Genkit.
• Language: TypeScript