ML Platform & AI Engineer
Sioux Falls, SD · open to remote
I build the systems around machine-learning models: the training, evaluation, and developer tooling that get them out of notebooks and into production. On my own time, that means self-hosted AI tools that only ship once they pass their tests.
Self-hosted LLM tooling, gated by evals.
An MCP server that gives AI assistants semantic search and note editing over a personal knowledge base, behind its own OAuth. Small local models answer only the kinds of question they have earned: 30+ labelled cases, 95% right, no flips across repeats. Runs on a Raspberry Pi k3s cluster and a home GPU.
A chess engine, forged in Rust.
A bitboard chess engine built for measured performance work. Its move generation, checked against perft, runs at about 52 million positions a second, and search runs at about 1.5 million nodes a second. A change to playing strength is kept only if an SPRT match shows it helps. It plays as a bot on Lichess, rated about 2000 in bullet and blitz.
One flake input, a full dev environment.
A Nix framework for reproducible development environments. Pin it as a flake input, call one function, and get the same setup on macOS, Linux, WSL2, or NixOS with a single command.
also: lex, an entropy-based Wordle solver in Rust.