Christien Antonio

Solo AI researcher building Aurelius — a 4B hybrid linear-attention (GDN 3:1) LLM. Mechanistic interpretability: attention circuits, key-collapse laws, output-gate write-path constraints. Certified-sparse attention + tile-ANS entropy codecs for LLM serving. Budget-constrained RLVR (SFT-first). Fleet: M1 Pro, Dell Precision 3581, Colab.