I build offline speech software for Apple computers. I founded Tenzan Logic, which makes Voxmill and Helio.
| Project | What it does |
|---|---|
| hdfour | A read-only HDF4 and HDF-EOS2 reader in Rust, with Python and WebAssembly bindings. Tests compare its output with the reference HDF4 C library. Work in progress; the bindings build from source. |
| BETON | A Claude Code plugin that runs installed linters, type checkers and selected tests after edits, then returns failures to Claude. A separate guard rejects common lint and test bypass markers. |
| gba | A Game Boy Advance emulator in C++23 and SDL3. Tests use programs written for the emulator, without distributing game ROMs. |
| Planetary System Simulation | A two-body gravity simulator using SciPy. Tests check energy and momentum conservation and compare the orbital period with Kepler's third law. |
Other projects: core-chat, a streaming chat client with tests that use a stand-in model; Fruitend, a CSS illustration with an element inspector; and Cook Book Chronicles, a recipe site with serving adjustments and cooking timers.
withastro/astro #18660, merged: preload variable fonts when the requested weight range is inside the file's range.
HSPMN: Python research code for an unpublished manuscript on hybrid language models. It compares a learned per-token gate with a frozen random gate for routing between a Gated DeltaNet recurrent stream and a Native Sparse Attention stream. I trained models on WikiText-103 at two scales, 140M and 350M (scale names, not exact parameter counts), on one RTX 5090.
A later audit found a bug that shifted the training targets twice, so every recorded run learned to predict the token two places ahead, not the next token. The audit also found future-token access in the predictive-coding gate and automatic NSA block selection, and an incorrect Gated DeltaNet CPU recurrence. The reference implementation in version 5.1.0 fixes these defects; I have not yet run it at training scale (reproduction status). Manuscript source (LaTeX), written before the audit.
- Voxmill: offline transcription for Apple computers, built with Rust, whisper.cpp and Core ML. 7.74% word error rate on the 758 read-speech clips of the Polish FLEURS test set (Voxmill Pro, Whisper large-v3-turbo Q5_0, normalised text, measured by Tenzan Logic, September 2026; method and results). Meetings and interviews have not been measured.
- Helio: a demo that estimates energy lost to solar-farm inverter outages and underperformance from the data the inverters already log. It uses public data; curtailment orders in the demo are simulated.

