LLMagic is a collection of small, visual, data-driven experiments that explore the internal behavior, structure, and variability of large language models (LLMs).
Each mini-project aims to reveal how models think, respond, and drift — through interactive plots, embeddings, and statistical analysis.