I developed a PyTorch version of the ShapeWorks optimizer as a sandbox environment to easily try out new loss functions and approaches for shape modeling. Using PyTorch's autograd function helps with not having to write gradient functions, which saves a lot of time. Furthermore, writing new conceptual routines in Python is much faster than adding code to the massive ShapeWorks back end. This code is used as a sandbox to explore new ideas.

I work at Scientific Computing and Imaging Institute on the ShapeWorks project. ShapeWorks is a free, open-source suite of software tools that uses a flexible method for automated construction of compact statistical landmark-based shape models of ensembles of anatomical shapes that does not rely on any specific surface parameterization. The method requires very little preprocessing or parameter tuning, and is applicable to a wide range of shape analysis problems, including nonmanifold surfaces and objects of arbitrary topology. ShapeWorks includes tools for preprocessing data, computing landmark-based shape models, and visualizing the results. 

In particular, I was in charge of implementing shape modeling for arbitrary regions of interest. This is a complex research problem with heavy development elements.

https://github.com/SCIInstitute/ShapeWorks 

I worked on automating the pipeline for single particle cryo-electron microscopy  grid screening using deep learning. We built a dataset with our collaborator and wrote Python Tkinter GUI to collect quality labels. Then we trained a quality assessment network to predict such labels for grid areas.

"Cryo-electron microscopy (cryo-EM) of single-particle specimens is used to determine the structure of proteins and macromolecular complexes without the need for crystals. Recent advances in detector technology and software algorithms now allow images of unprecedented quality to be recorded and structures to be determined at near-atomic resolution. However, compared with X-ray crystallography, cryo-EM is a young technique with distinct challenges." [Cheng 2016]

As a personal project, I built a automatic software to play the desktop version of the game CookieClicker. I used pynput to simulate clicks on the screen with set routines to play the game automatically. I also attempted to use CV2 template matching tools to recognize images, but it was too slow to work in real time.

https://github.com/HeavenlyBerserker/AutoCookieClicker.git

I worked on a proprietary project for Exxon Mobile, which consisted of writing CUDA code to accelerate research code to enable the viability of modeling geological features.

Back when VR was a novelty, I decided to create a VR lego building experience as my senior project. I used Google Sketchup to build the pieces. Then used Unity and C# to write the logic and connect it to the HTC Vibe.

https://github.com/HeavenlyBerserker/Lego-VR.git