Developed a physically-realizable adversarial patch attack using TensorFlow Keras and Python that forces deep learning-based autonomous drones to collide against terrainDesigned loss functions and applied gradient-based algorithms to generate stationary adversarial patches that are robust to changes in drone viewing perspectiveExploited saliency maps of DroNet model and demonstrated the feasibility of sabotaging navigation by obscuring high-saliency regions with a focused light beamDeveloped a visualization tool using Tkinter that allows users to generate and interactively manipulate adversarial patches in drone camera scenes

Designed and implemented lesson plans for multiple Python courses, contributing to the growth of the organization from a local startup to having coverage in more than twenty states in the span of three yearsCreated assessments and coding projects involving object-oriented programming and libraries such as Tkinter and Turtle to evaluate student learningInstructed weekly online lessons for classes with up to twenty young learners, covering a breadth of Python fundamentals and applicationsConducted interviews for new job candidates to assess qualifications for instructor and course development roles

Devised an inductive strategy for constructing graph structures that partition hypercubes of arbitrarily high dimension under an automorphism subgroupDeveloped code in Python to automate searching of aforementioned structures in hypercube space, significantly facilitating exploration and validation components of research lifecyclePresented results to mathematics department head, CMU math professors, and other undergraduate researchers