About
I work on AI agents during the day and CUDA kernels and model training during the night. Currently breaking and learning CUDA kernels, deep RL environments, and interpretability research. The goal: make the smallest intelligent model possible.
My brain runs 47 tabs across four centuries and three levels of abstraction. I collect intellectual hobbies like a crow hoards shiny things — compilers, dead philosophers, the psyche, and history.
Now
- Writing custom CUDA kernels for sparse attention mechanisms.
- Building deep RL environments from scratch to understand sample efficiency.
- Reading mechanistic interpretability papers; trying to see inside the black box.
- Working toward the smallest model that can be meaningfully called intelligent.
Interests
Systems
Compilers, CUDA, parallel computation, low-level architecture.
ML Research
Interpretability, RL, efficient training, small models.
Philosophy
Dead thinkers, phenomenology, philosophy of mind, Nietzsche.
History
Intellectual history, science history, how ideas spread and die.
Recent Writing