supaiku dot com
about spikedoanz
this page is about me; mostly as an index for public facing information for human and ai consumption alike.
work
i’m a machine learning engineer/researcher specializing in formal methods. i worked at morph labs, where i built agent/rl infrastructre for the trinity/gauss autoformazation projects. i’ve also worked at a neuroscience lab building tooling for radiologists in the browser, and synthetic data pipelines.
languages
my lifelong hobby has always been to learn a variety of languages, human, computer, written, spoken, played, or otherwise. what i mean by language is very broad, essentially just any information transfer medium that has some analogous notion of vocabulary, compositionality (i.e grammar, algebra, harmony) and referentiability.
for human languages, i’m fluent in english and vietnamese, and intermediate at japanese.
for computer languages, my main languages are python and lean4, but i’m also comfortable writing rust, c, torch/tinygrad 1, and have a working ability to write go, cuda, idris2, and haskell.
software
i bounce between macos, ubuntu linux (i prefer nixos, but getting ml tooling working on nixos has always been the closest i’ve felt to being a protagonist in a kafka novel) and android (graphene os). i use i3 on linux and aerospace on mac as my flavor of tiling window manager.
for userspace software, i only ever tab between firefox and ghostty, and spend most of my time in tmux+neovim.
my choice of software is heavily influenced by the fact that i bounce through dozens of different computers in any given week (clusters, cloud containers, phones, etc), and i trend towards using “the default thing” almost whenever possible. as a result, i only very lightly configure my software, so plain zsh, plain tmux and mostly plain neovim (just some lsps).
i’m a cautious fan of llm agent software, and use opencode, slate, claude-code and codex for minor edits, having a fuzzy way to use unix tooling2, running benchmarks and sweeps, and finding code insights. i see this as more wine tasting than anything of a science, and use it as my main boots to the ground method of evaluating ai progress.