Research

Compilers for scientific computing

Alongside the libraries and the platform, SGCX is building programming languages for numerical work. They follow the same principles as everything else here: correctness is checked independently, nothing falls back silently, and the programmer stays in explicit control of where and how code runs. Both projects are experimental and not yet publicly released.

Experimental

DVEB

DVEB is an experimental scientific-programming language and ahead-of-time compiler for Linux and NVIDIA GPUs. It compiles scientific programs into native CPU and CUDA code, giving researchers explicit control over memory, parallel execution, and data movement while reducing the need for handwritten C++/CUDA infrastructure.

Development is driven by real scientific workloads, including fluid dynamics, maximum-likelihood estimation, and time-series models. Each capability is evaluated through independent numerical checks and reproducible benchmarks, with an emphasis on correctness, performance, and compact native deployment.

Experimental

Alloy

Alloy is an ahead-of-time compiled language for numerical computing on Apple silicon. A kernel is written once, and the compiler produces both native ARM CPU code and a precompiled Metal GPU kernel, packaged together in a single bundle.

At runtime the caller chooses the target explicitly — CPU, GPU, or automatic — and the runtime reports what ran and why. Nothing falls back silently, no shaders are compiled at load time, and CPU and GPU share the same memory without copies.

Open Source

Our statistical libraries, PyStatistics and PyStatsBio, are freely available and we welcome contributions.

View on GitHub →