independent researcher

Arabic language, model evaluation,
and inference systems.

Exploring Arabic NLP, model evaluation, and representation analysis through rigorous, reproducible research.

read the research notes

projects

ember

Inspectable CPU inference for reproducible LLM systems research.

view project
01

atlas

Maps limitations, missing evaluations, and future-work claims from papers. Deterministic. Evidence-first.

view project
02

mini-os

A minimal x86_64 kernel in Rust with paging, async execution, ramdisk, and a small shell.

view source
03

research notes

The File Said Q4. The Runtime Said f32.Ember    runtime systems Reproducibility Starts Before Inferenceevaluation    reproducibility Why Research Debt Atlas Does Not Generate Research Gapstooling    reproducibility When ArabicMMLU Accuracy Depends on Answer PositionArabicMMLU    evaluation When the Result Gets Less Flashy but More Realprobing    methodology
view all notes