I'm a recent graduate with a Master's degree in Computational Engineering from Mississippi State University with a passion for all things technically challenging. With a strong foundation in research and problem-solving in the area of big data and machine learning, I’m making my transition into professional software development through personal projects and continuous learning of state-of-the-art techniques in both areas and more.

• 🎓 Academic background in Applied Mathematics and Scientific Computing
• 💡 Interested in Deep Learning, Scientific Computing, Data Science, Computer Vision, High-Performance Computing (HPC), and Machine Learning in general
• 🌱 Currently learning about Vector Databases, RAG Pipelines, and the newest MLOps techniques
• 💞️ I’m looking to collaborate on any and all cool projects so I can keep developing myself as a programmer, mathematician, and ML engineer while getting better at DevOps

My Github may look a little sparse because for a long time, I've been bad about using git in smaller projects meant only for myself. Also, as my focus became more mathematically-oriented, the code for many projects often included mostly simple scripts and quick calculations. I'm getting better with DevOps now and intend to push many old projects when time permits. Many recent projects you see on my profile are those that began some time ago, but were never organized into a true code repository.

• A Python module implementing "CAP-VSTNet: Content Affinity Preserved Versatile Style Transfer" that allows for modular integration of style transfer as an image augmentation in deep learning pipelines, with a fully Pytorch-based framework for image and video training and inference.
• Supports image/video style transfer, segmentation mask guidance, multi-style interpolation, and high parallelization for deep learning workflows
• NOTE: The main branch is currently behind a separate branch, which implements the full Python module implementation with more proper documentation and more flexible implementation with data validation through Pydantic.

• A modular, multi-mode turret platform combining motion detection, object detection using pre-trained deep learning models, real-time control, and physical actuation to detect and deter pets (e.g. cats) from tampering with hazardous houseplants using an automated water jet.
• A client-server model connects a Raspberry Pi client and desktop Flask server running CUDA for real time detection and turret operation.

• An efficient, interactive GUI framework for reviewing and categorizing image annotations and more, supporting single-label and multi-label classification with a results (slideshow) viewer for quick inspection and optional generation of more advanced views
• Allows generation of images with on-the-fly overlays (bounding boxes, segmentation masks) and slideshow inspection, accelerating manual QA with a low code interface

• A lightweight Python-based security system that monitors a webcam for motion, records short video clips upon detection, encrypts them with AES-GCM, and automatically uploads to a cloud backend (Cloudinary or Dropbox).
• It's a simple and easy solution to catching burglars' faces for police even in the event your computer is stolen. It's designed for minimal dependencies and easy extension to native-code capture modules.
• The backend for video monitoring and motion detection uses OpenCV while a producer-consumer design uses multi-threaded event loops for logging, encryption, and publishing videos.

• Created a powerful NLP pipeline structure for automatically labeling and sorting bookmarks and exported URLs (e.g. from OneTab or browser bookmarks)
• Enabled extracting link metadata, scraping webpages, and applying advanced summarization, embedding, and clustering to sort and label bookmarks
• NOTE: currently in a less-than-ideal state due to the need to revisit the tokenization, filtering, and embedding logic of final scraped webpage data

• A simple CLI tool for reformatting poorly-cropped webtoon-style pages (continuous vertical comic strips) into more intelligently stacked images, with the ability to export the new images to cbz, cbr, and pdf files
• Provides dynamic vertical segmentation, resizing, and padding by detecting blank horizontal regions (those between panels) with low variance to avoid cuts at panels or dialogue bubbles
• Efficient data streaming, dimensionality and image file type handling via metadata, and segment-based reconstruction allows for efficient creation of resized chapters

• !! EARLY STAGES !!
• implementation of a set of novel Segformer neural network variants dubbed "Lipschitz-Regularized SegFormer (LRSegformer)," which includes Lipschitz-regularized MLP decoder layers to improve model robustness, generalization, and stability.
• The 4 variants (types of regularization in the MLP layers) to be tested include the learned geometric mean (used in my thesis), a stable softplus-based regularization, a spectral normalization approach, and regularization by enforcement of weight orthogonality
• This repo is structured as an intended set of experiments for evaluating LRSegformer's robustness/generalization capabilities

• !! EARLY STAGES !!
• A modular Python library for evaluating, analyzing, and testing feature representations extracted from deep neural networks - meant to encapsulate most of the evaluation code used in my thesis in a clean, modular structure
• Designed with flexibility in mind, this toolkit supports evaluation workflows that rely on precomputed features or real-time streaming of probabilities/logits and features from external pipelines.

• !! EARLY STAGES !!
• my attempt at creating a whole engine (generator, solver, validator, and GUI) for a type of logic puzzle that I was introduced to recently; may not go much farther than a novelty tbqh
• eventually hoped to implement the underlying algorithms in C (already started) and create a PR for Simon Tatham's puzzle collection

• !! EARLY STAGES !!
• a modular refactoring of a core component of my thesis project: high-level wrappers for a wide range of image-based augmentations, allowing for built-in hyperparameter optimization to automatically tune augmentation application probabilities and parameter values

• !! EARLY STAGES !!
• A lightweight tool for parsing job board postings to extract and categorize technical skills and related metadata using LLMs and traditional NLP approaches

• !! EARLY STAGES !!
• planned to be an AI‑powered knowledge explorer designed to turn an unstructured library of research papers, e‑books, lecture notes, and internal documents into an interactive, searchable knowledge graph
• We'll use a modern Retrieval‑Augmented Generation (GraphRAG) pipeline with concept‑mapping and summarization agents so that users can use natural‑language queries to retrieve and summarize documents while enabling similar concept mapping.

! TODO: NEEDS UPDATING

Languages: Python | Java | C | C++ | MATLAB | R | C# | Kotlin | Javascript | Julia | Bash | SQL | Fortran

Frameworks & Libraries: PyTorch | scikit-learn | Tensorflow | CUDA | OpenCV | Jax | MxNet | Raytune | XGBoost | Numpy/Scipy | Pandas | Polars | Matplotlib/Seaborn | Plotly

Tools & Platforms: Git | Docker | CircleCI | Github Actions | git-lfs | Neptune | Tensorboard | Robot Operating System (ROS)