Selected work
A curated subset of projects, research software, and open-source contributions from my GitHub.
Featured projects
tinyflow
From-scratch flow matching experiments in tinygrad, including ODE solvers, training loops, path definitions, and a more modular image-generation setup.
attention
A transformer playground in PyTorch with standard attention, flash-attention experiments, and tokenizer/model variations for small language-model style training.
ParameterEstimation.jl
A Julia package for estimating parameters and initial conditions of ODE models from measurement data, developed as part of my PhD work.
StructuralIdentifiability.jl
Research software in the SciML ecosystem focused on structural identifiability algorithms for differential equation models.
Structural Identifiability Toolbox
An online tool I built and maintained for assessing parameter identifiability in dynamical systems.
Other personal projects
bpe
A byte-pair encoding implementation in C, built to study tokenizer internals and low-level data structures.
numpy-learn
A small deep learning library in NumPy, inspired by the PyTorch style of building models and training loops.
tasklist
A Rust CLI experiment for managing tasks with a simple Kanban-like board view.
gitignore-create
A CLI tool for generating .gitignore files quickly, installable via pip install gitignore-create.
Open-source contributions
llama.cpp
Contributed a ConvTranspose2D implementation to llama.cpp.
Symbolics.jl
Added a degree function implementation to Symbolics.jl. View pull request
SIAN-Julia
Maintained the package during my PhD work. Development fork
For a broader list of experiments and repositories, see my GitHub profile.