Hardware

My primary research testbed consists of NVIDIA Jetson Orin Nano and Jetson AGX Xavier boards — these are the devices I profile and benchmark for all my edge AI inference work. For model training and large-scale simulations, I use a custom workstation with an RTX 3090 and 64GB RAM running Ubuntu 22.04 LTS. On the go, I use a MacBook Pro. I keep a laptop stand to bring the screen to eye level, which makes a noticeable difference over long working sessions.

Editor & IDE

I use VS Code for most of my work — large Python projects, Jupyter notebooks, remote SSH into the Jetson cluster, and Docker-based workflows. For quick config edits and remote terminal sessions, I fall back to Neovim.

Essential VS Code extensions I rely on: Python, Pylance, C/C++, Remote - SSH, Docker, GitLens, Jupyter, and GitHub Copilot. For ONNX model inspection I use Netron.

Terminal & Shell

I run Alacritty as my terminal — it's GPU-accelerated and noticeably snappier than the alternatives. For session persistence and multi-pane layouts, I use tmux. When running many experiments in parallel, I launch a tmux session with multiple panes from a shell script.

My shell is Zsh with Oh My Zsh, the Powerlevel10k theme, and plugins for autosuggestions and syntax highlighting. I've set unlimited history size — being able to search through old commands with Ctrl+R has saved me more time than I'd like to admit.

Research & ML Stack

For deep learning, my primary stack is PyTorch for training and prototyping, ONNX Runtime for portable inference, and TensorRT for optimized deployment on Jetson hardware. I convert most models to ONNX format before deploying to the edge cluster — it makes profiling and benchmarking significantly more reproducible.

For computer architecture and network simulation, I use SST Core / Elements and NS-3. These are essential for the performance modelling work — evaluating latency and throughput characteristics under concurrent workloads before deploying to real hardware.

For containerization, I rely on Docker and Docker Compose for local development, and K3s (lightweight Kubernetes) for managing the edge cluster. Data analysis and visualization happens in NumPy, Pandas, Matplotlib, and Seaborn — nothing exotic, but reliable and fast for generating publication-quality figures.

Writing & Productivity

For paper writing, I use Overleaf exclusively — the real-time collaboration with co-authors is indispensable. I manage references with Zotero and its browser connector. Diagrams and system architecture figures go through draw.io.

For general note-taking and project management, I use Notion. Day-to-day communication with the team happens on Slack. All code lives in GitLab (for UGent work) or GitHub (for personal projects).

Version Control

I use Git for everything. For large research projects with multiple experiments running simultaneously, I've found git worktrees invaluable — each experiment branch gets its own working directory without duplicating the entire repo. For commit messages, I keep them descriptive but don't overthink it; the experiment logs in the repo are what matters for reproducibility.

Not every item on this page is updated constantly, but if something is listed here, I've used it in the past year and was happy with it. This is a snapshot of my personal, opinionated setup — what works for me might not work for everyone.