I started studying mechanistic interpretability in 2026. I’m pursuing it full time, focusing on how faithfully interpretability tools reflect model internals. My work covers model weight structure, MoE routing, and Jacobian lenses. I also work as a full-stack developer for a nonprofit.
This account hosts my research releases. My earlier research and software projects are on my other account.
Latest release: Repeated blocks in Qwen-family checkpoints—a reproducible weight audit of 33 Qwen and Qwen-derived checkpoints, ranging from 0.6B to 2.4T, plus SOLAR as a calibration.
Qwen3.6-27B and J++ readout check · Hugging Face
Self-funded, with no institutional affiliation.