This project implements Andrej Karpathy’s nanoGPT architecture entirely within a spreadsheet, replacing Python with Excel/Google Sheets formulas. It encodes all transformer components—embedding, layer normalization, self-attention (KQV), projection, MLP, softmax, and logits—as interactive matrix calculations. The project serves as a visual learning tool, allowing users to manipulate and trace data flow through the entire character-based prediction system, which uses roughly 85,000 parameters configured for characters A/B/C.