About

One engineering problem

I work across machine learning, language-model architectures, operating systems, compilers, and experimental software. I am especially interested in the places where these areas begin to overlap.

I learned much of this by building: kernels, parsers, training pipelines, model architectures, APIs, and products.

How I work

Understand the mechanism. Build the smallest real version. Measure it. Find what is wrong. Iterate.

I started on low-end hardware with limited compute and connectivity. That pushed me toward understanding systems directly and making experiments fit the hardware I actually had.

A significant part of my research is designing experiments that can be reproduced under constrained compute. Careful baselines and honest limits matter as much as the mechanism.

Current research

Alethic explores Sin-Gated Control Attention and controlled comparisons with conventional transformer baselines. The current Phase-I direction targets approximately 1B parameters with an 8,240-token context.

Systems and languages

AiOS explores boot, kernel components, memory management, and later Linux/rootfs integration. Flang-Lite connects a lexer and parser with an AST, NASM code generation, and a bytecode VM.

Research progression

1. Small-model experimentsSinGatedLM and preliminary parameter-matched comparisons.
2. Model pretraining116M predecessor and Alethic-151M training.
3. Current directionSGCA, controlled validation, and Phase-I preparation.

A progression of work; exact project start dates are not assigned where they were not supplied.

Beyond the current run

I’m a self-directed secondary-school student preparing for university. My longer-term interest is understanding intelligent systems across models, languages, operating systems, and computer architecture.