Flow is a native programming language and ecosystem for concise, predictable high-performance software. It is where much of my work in compilers, numerical computing, graphics, DSP and AI-assisted development is converging.

flow-scikit — scikit-learn reimplemented in Flow, compiling to a native binary without a Python runtime · doom-flow — DOOM ported to Flow · flow-euler — Project Euler as a language/compiler test suite

My AI work is distinct from my general software work: models, agents, learning systems, machine intelligence infrastructure and AI-native tooling.

tinyML — real-time C++ machine learning and statistical analysis · Fractal-Memory-Network — agentic memory experiments · safe-yolo-agent — allowlist-based permissions for autonomous coding agents · CycleGAN Timbre Transfer — neural audio timbre transfer

Earlier work includes neural networks from scratch, genetic algorithms, stereo disparity networks, OCR, stance detection, evolutionary agents, Slow Feature Analysis, NEAT and HyperNEAT.

Flow — programming language and compiler ecosystem · azazel — deterministic Zig/CUE build system · zaza — build system for C, C++, Zig, CMake and WebAssembly · pypi-toolkit — Python packaging automation

Audio software has been a continuous thread through my work: DSP, synthesis, effects, plugin architectures, audio engines and ML for sound.

danzig — VST3 framework in pure Zig · Analog-Fattener — JUCE audio effect · JUCE Ladder Filter — DSP implementation · isobarcpp — algorithmic music tooling in C++

Professional audio and product work lives primarily through Quilio.

ShaderLibrary — GLSL, Metal and Slang shaders · voxel-planet-demo — voxel planet renderer · Hydra-Sketches — live-coded GPU graphics · SocietySimulator — computational society simulation · Decomposition — top-down game

My research background is in neuroscience, with current work extending into dynamical systems, computational modelling, scientific infrastructure and tools for interrogating research itself.

perturbation-kernel — scalar, SIMD and GPU perturbation-kernel estimators · citeverify — bibliography verification against OpenAlex, Crossref and arXiv · corpus-lens — corpus analysis, similarity and embeddings · refereed — multi-perspective computational research review

How much of the stack can be made simpler when you're willing to reconsider the abstraction itself?

That question has taken me through audio DSP, neuroscience, machine learning, programming languages, build systems, graphics, games and scientific computing. Flow is increasingly where those threads meet.