This project provides a complete implementation of a transformer-based language model in pure Rust without external ML frameworks. It includes pre-training on factual text completion, instruction tuning for conversational AI, and an interactive chat mode. The implementation covers full backpropagation with gradient clipping and a modular transformer architecture built from embeddings, transformer blocks, and output projection layers.