Hello! 👋

I'm Josh Euikwang Josh

Machine Learning Engineer

About Me

I'm a MSE CIS student at the University of Pennsylvania (GPA: 3.90) with a concentration in Artificial Intelligence. I finished my undergrad at Johns Hopkins University as a Computer Science and Neuroscience double major (GPA: 3.70). I have professional working experience as an R&D Engineer at Synopsys and an NLP Researcher at NCSoft. I currently serve as a Teaching Assistant for CIS 5200 (Machine Learning) at Penn and am a Research Assistant at the Physics-Informed Machine Intelligence Lab. Besides work, I enjoy performing and playing tennis. I am currently a tenor for the Penn Counterparts, UPenn's Oldest All-gender A Cappella. I was also a tenor for The AllNighters, JHU's Award-Winning All-Male A Cappella.

Python C++ PyTorch OpenGL AWS Docker RoBERTa OpenAI API CUDA Jenkins GitLab CI MongoDB React Node.js SLURM

Education

M.S.E. in Computer & Information Science

University of Pennsylvania

2025 - 2027
  • Artificial Intelligence Concentration
  • Cumulative GPA: 3.90 / 4.0
  • Teaching Assistant: Machine Learning (CIS 5200)
  • Penn Counterparts, UPenn's Oldest All-gender A Cappella

B.S. in Computer Science, Double Major in Neuroscience

Johns Hopkins University

2019 - 2023
  • Cumulative GPA: 3.70 / 4.0
  • Dean's List all semesters (2019 – 2023)
  • Winner of Best Automation Application, HackViolet 2021
  • Nu Rho Psi Neuroscience Honor Society
  • The AllNighters, JHU's Award-Winning All-Male A Cappella
  • Brazilian Jiu-Jitsu Club

Experience

Teaching Assistant — Machine Learning (CIS 5200)

University of Pennsylvania

Aug 2025 - Present
  • Teaching assistant for CIS 5200 (Machine Learning), supporting instruction for graduate and advanced undergraduate students.

Research Assistant

Physics-Informed Machine Intelligence Lab

Sep 2025 - Present
  • Conducting research in physics-informed machine learning at the University of Pennsylvania.

R&D Engineer

Synopsys

Mar 2024 - Jun 2025
  • Spinnaker: Integrated off-the-shelf Physics-Informed Neural Network (PINN) support into a Python framework for scalable, physics-informed ML for TCAD.
  • Developed multi-node, multi-GPU training compatible with torchrun and SLURM; accelerated training by up to 59.5% through CUDA Graph integration.
  • Implemented a self-adaptive PDE loss weighting algorithm to improve PINN convergence and usability.
  • Built a CI/CD pipeline using Jenkins, Perforce, Docker, and coverage.py for automated testing and code coverage tracking.
  • Process Explorer: Built a Python-based GUI test interface using Qt, Squish, and MongoDB for a hyper-realistic 3D semiconductor process emulator.
  • Created 100+ automated GUI tests achieving 80.5% line coverage; set up CI/CD with GitLab CI, Docker, Gcov, and Coverity to automate testing and memory leak checks.
  • Led a C++ Fundamentals Seminar (Apr 2024 – Jul 2024) for internal engineers.

NLP Researcher

NCSoft

Jul 2023 - Jan 2024
  • ATOMIC KR: Implemented symbolic knowledge distillation to computationally generate a Korean relational commonsense knowledge graph dataset of 600,000+ triples.
  • Developed a Python module to asynchronously call the OpenAI API, significantly optimizing symbolic knowledge distillation speed.
  • Built a RoBERTa-based critic model and custom annotation platform enabling 5 people to grade 10,000 data points in 7 days, vs. 150 people grading 3,000 in the original study.
  • KCoT: Developed a Python text processing module to process 20+ years of data from 16 Korean national exams for manual data annotators.

Research Assistant

Janice Chen Lab, Johns Hopkins University

Aug 2022 - May 2023
  • Developed a Python tool to convert Universal Sentence Encoder embeddings of fMRI audio narratives into event networks, analyzing semantic links' effects on memory recall.
  • Engineered a parallelized audio transcription module with OpenAI Whisper and Joblib, boosting transcription speed by 400%.

Research Assistant

Isik Computational Neuroscience Lab, Johns Hopkins University

May 2021 - May 2022
  • Analyzed the influence of biomechanical motion on social perception by correlating AlexNet embeddings from naturalistic videos and their motion-only counterparts with sociality ratings.
  • Curated and enhanced the quality of crude OpenPose-filtered motion video stimuli through smoothing.

Projects

01

Reinforcement Learning for Claim Verification with Grounded Justification

Fine-tuned GPT-4.1-nano via SFT and DPO on the FEVER benchmark using an RLAIF teacher–student pipeline, achieving 0.928 macro accuracy on a ternary fact-verification task. Engineered and benchmarked three justification styles across 14 training configurations, with evidence-first grounding yielding the best results by anchoring model reasoning in source passages.

GPT-4.1SFTDPORLAIFPython
02

Minecraft Clone (Mycraft)

Built a multithreaded Minecraft clone in C++ with OpenGL featuring mutex-protected worker threads for block/VBO data, third-person character system with a rigged humanoid player model, zombie NPC AI, distance culling, and physics-based collision detection.

C++OpenGLMultithreading
03

Diagnosing Parkinson's Disease from PET Scans using 2D-CNNs

Developed an efficient 2D-CNN to detect Parkinson's disease from 3D PET scan slices with 90.3% accuracy. Enhanced model interpretability with GradCam, aligning model predictions with clinical expectations.

PyTorchCNNGradCamPython
04

Iieum Tutoring

Developed the relational models and launched a full-stack MERN tutoring platform on AWS Lightsail. Implemented user authentication, site navigation, and a dynamic class scheduling system.

MongoDBExpressReactNode.jsAWS