building ML systems
from first principles.
I implement ML infrastructure from the ground up - currently a Rust autograd engine - and work as an AI model trainer applying RLHF to improve model reasoning. PSU Mathematics, 2023. I care about understanding how things actually work.
Tensor library and autograd engine in Rust. Implements forward and backward passes without ML frameworks - similar scope to Karpathy's micrograd but systems-level. Goal: a solid foundation to build a training loop from scratch.
RAG-based chatbot that answers Neovim configuration questions by indexing plugin documentation and community knowledge. Built to scratch my own itch - Neovim config rabbit holes are deep.
I'm a software developer and mathematician focused on ML systems. B.S. in Mathematics from Penn State - coursework in real analysis, abstract algebra, probability theory, linear algebra, and numerical analysis.
Currently contracting as an AI model trainer, doing RLHF on LLM outputs in programming and mathematical reasoning. Writing code since I was 12. Still doing it for the same reason: I want to understand how things work.
Open to ML engineering and AI roles. For opportunities or an interesting conversation:
richard.rj.larson@gmail.com