Projects

Open Source Contributions

56+ merged PRs across 15+ repositories. Highlights below — full history on GitHub.

Shells & CLI tools (Rust)

• Nushell — 18 merged PRs. Ported uutils commands (cp, mv, ln, uname) into Nushell as drop-in replacements, coordinating changes across repos. Bug fixes, feature additions, and docs.
• uutils Coreutils — 12 merged PRs. Refactored commands into public APIs for Nushell consumption. Bug fixes to chmod, cp, ls, mktemp, hashsum.
• just — Tempdir configuration.

Python-Rust interop

• PyO3 — Added num-rational support for fractions.Fraction, Decimal scientific notation, PyPy compatibility fix.
• maturin — Added uv as develop backend.

Rust tooling

• cargo-semver-checks — Wrote 3 new semver lints for checking semver violations, and added cli comma-0separated feature flag support.

Terminal & desktop (Zig)

• Ghostty — Nushell shell/SSH integration, GTK tab title, i18n for Nautilus script.

Data science & visualization (Python)

• Matplotlib — 8 merged PRs. Docs, examples, TextBox widget tests, bug fix for empty data plotting.
• YOLOv5 — Disallow --dynamic when --half is passed.

Other Projects

Website for NL Organization

• Developed backend/frontend for the website
• Member authentication for downloading private organization files

Check out the website

KITTI Data to ROS bag

• Tool to convert raw data from the automotive KITTI dataset to a ROS bag, using the ROS PointCloudLibrary

Github

Deep Learning Hobby Projects

Older projects from when I was exploring ML/computer vision.

Kaggle Xray Detection

• Kaggle competition to detect 14 different chest x-ray anomalies
• Trained EfficientDet network, evaluated using COCO evaluation
• Deployed locally using Flask

Github

PointRCNN Detections with ROS

• Implementation of state-of-the-art object detection on the KITTI dataset with ROS
• Detections visualized in LiDAR and camera space using RVIZ

Github

Plant Instance Segmentation

• Implemented Mask-RCNN for counting and segmenting plants on a private dataset
• Evaluated qualitatively and quantitatively using COCO evaluation
• Trained with Google Cloud Platform

CityScapes Semantic Segmentation

• Multiple experiments with optimizers, regularization, and input variations
• Google Cloud training on CityScapes dataset

Robust Semantic Segmentation

• Used the Synthia Dataset to test whether data augmentation techniques (gamma, flip, saturation) improved network performance
• Google Cloud training on Synthia dataset

Kaggle Speech Recognition

• Classifying 1-second audio files using CNN with spectrograms as input
• Submitted to the Kaggle speech recognition challenge