Resume

• Architected Alpine, a modular PyTorch library to streamline the development of deep learning model development and training.
• Engineered a neural network pretraining method, boosting PSNR by 25-30 dB over meta-learned baselines, with 2x faster convergence.

• Developed a high-speed robotic vision system using ROS and OpenCV to track rocket trajectories in high-speed video.
• Pioneered a hybrid Neural-PID controller, decreasing tracking error by 94% and selected for deployment in Artemis moon missions.

• Automated the validation pipeline for RF field stabilization, eliminating the need for $350k in legacy hardware resources.
• Replaced manual PID tuning with robust self-tuning control algorithms for precise temperature control in particle accelerators.

• Collaborated with DoE researchers to achieve a 92% runtime improvement in MPAS Ocean through tracer advection modeling.
• Engineered GPU/TPU-compatible heat transfer models by reformulating physics equations into efficient tensor algebra operations.

Vyas, K., Kayabasi, A., Kim, D., et al. (2025).

The Surprising Effectiveness of Noise Pretraining for Implicit Neural Representations. Submitted to IEEE/CVF CVPR 2026.

Kim, D. (2024).

Engineered Underwater Vehicle for Ocean Litter Mapping. New Mexico Journal of Science.

Kim, D., & Iturregui, A. (2022).

Developing a Control Algorithm and Simulation for Thrust Vector Controlled Rockets. New Mexico Journal of Science.

• GPA: 4.0/4.0
• Scholarships: Max Roy Scholarship (Full Tuition), Los Alamos Lab Gold Scholarship ($20k)
• Coursework: Data Structures & Algorithms, Computer Vision, Machine Learning, OOP, Signals & Systems, Linear Algebra

• Founded a real-time productivity platform with AI-driven study assistance and focus environments (Dart, Google Cloud).
• Boosted system reliability for 1200+ users through optimized indexing, secure user authentication, and CI/CD.

• Developed AUV to capture subsea footage and process it with YOLOv5-based object detection (C++, Python).
• Implemented deep learning and SLAM-based mapping to discover 174,000+ pieces of ocean trash (Regeneron ISEF 3rd Place).

• Simulated over 8000 active satellite trajectories using OpenGL, ImGui, and C++.
• Implemented a modular graphics pipeline with optimized memory usage for smooth 3D visualization.