Building LLM Primitives from Scratch in TypeScript: Tokenizers, Vectors, and Attention

Products like ChatGPT and Claude can make LLM primitives look like a black box. Writing the underlying pieces as ordinary code exposes their intermediate data and operations. This tutorial builds three of them from scratch in TypeScript: a byte-pair encoding (BPE) tokenizer, cosine similarity vector search, and scaled dot-product attention.