independent ML systems research
voidwest
systems, ml, and the space where they overlap.
ml systems model internals arabic nlp inference tooling
github linkedin email ember research
current focus
ember cpu inference, probing, and validation workflows research notes arabic morphology, tokenization, and model internals tooling small local systems for experiments, probes, and reproducible artifacts

ember

ember
research layer for hidden-state extraction, leakage-aware probing, and reproducible morphology experiments over GGUF models, with a native rust backend for inspectability and llama.cpp as the scalable execution path.
rustinferenceggufcpuprobingvalidation
source research engineering internals

research notes

research notes
running notes on papers, experiments, and ideas. current focus: Arabic nlp, tokenization, and mechanistic interpretability.
papersnotesarabic nlpprobing

other projects

mini-os
minimal x86_64 kernel in rust. gdt, idt, paging, heap with 3 allocation strategies, cooperative async executor, ramdisk, and a shell with 10 commands. runs in qemu.
rustkernelx86_64no_std
devdocs-v2
local rag pipeline for querying pdf documents. sentence-transformers for embeddings, chromadb for vector storage, ollama for inference. fastapi backend with a vanilla html/css/js chat ui. fully local, no external apis.
pythonragchromadbollamafastapi