Experience

VoyagerX

• South Korea
• IT Services and IT Consulting
• 1 - 100 Employee

• Machine Learning Engineer

  • Aug 2021 - Present

  - Alternative military service as an IT skilled personnel for 2 years- Currently working on training large scale language model - Korean version of GPT-3- Implemented Morpheme-aware Byte-level BPE, a specialized Korean tokenizer, which instantly improved all metrics by ~30%- Implemented finetune pipeline on a downstream task (NER) using DeepSpeed - Developed a core scanning model for vFlat, which is used for inference 40M+ times by 2M+ users, monthly- Constructed a novel automated data labelling process while building the scanning model, which saved $300,000 for the company- Established a conversion process that enabled research in Pytorch and deployment in Tensorflow- Utilized: Python, Pytorch, DeepSpeed, Docker, Tensorflow, OpenCV, Numpy, PIL, Matplotlib, PyQt, Git, Anaconda

• Machine Learning Engineer Intern

  • Dec 2020 - Jul 2021

  - Worked on a novel deep-learning based scanner application, vFlat- Developed a core scanning deep-learning model, handling the entire deep-learning model development pipeline: from data generation & cleaning, to training & deployment- Developed a VGG-style CNN that detects finger location that helps providing better scanning experience with inference time limit ( < 15ms) considered on mobile environment- Utilized: Python, Pytorch, Tensorflow, OpenCV, Numpy, PIL, Matplotlib, PyQt, Git, Anaconda



Qualcomm

• United States
• Telecommunications
• 700 & Above Employee

• AI & DSP Developer Intern

  • Sep 2020 - Dec 2020

  - Worked on deep-learning based optical flow prediction and frame rate conversion.- Designed, implemented, and validated 10+ CNN architectures for optical flow generation.- Applied channel-wise and spatial attention, which significantly improved optical flow output- Utilized: Python, C/C++, Tensorflow, Convolutional Neural Network, Git, Anaconda - Worked on deep-learning based optical flow prediction and frame rate conversion.- Designed, implemented, and validated 10+ CNN architectures for optical flow generation.- Applied channel-wise and spatial attention, which significantly improved optical flow output- Utilized: Python, C/C++, Tensorflow, Convolutional Neural Network, Git, Anaconda



Korea University

• South Korea
• Higher Education
• 700 & Above Employee

• Research Intern

  • Jul 2020 - Aug 2020

  Graph Neural NetworkMachine Learning and Vision LabSupervised by Prof. Hyunwoo Kim- Conducted a study on machine learning with graphs (CS224W - Stanford Online Course).- Implemented several GNN architectures and undertook various experiments that combined CNN and GNN to enhance CNN’s locality and improve global inference.- Utilized: Python, Pytorch, Convolutional Neural Network, Graph Neural Network Graph Neural NetworkMachine Learning and Vision LabSupervised by Prof. Hyunwoo Kim- Conducted a study on machine learning with graphs (CS224W - Stanford Online Course).- Implemented several GNN architectures and undertook various experiments that combined CNN and GNN to enhance CNN’s locality and improve global inference.- Utilized: Python, Pytorch, Convolutional Neural Network, Graph Neural Network



Seoul National University

• South Korea
• Education Management
• 200 - 300 Employee

• Research Assistant

  • May 2019 - Aug 2019

  Autonomous vehicle, 3D SLAM, shell scriptingReal-time UBIquitous System Laboratory (RUBIS)Supervised by Prof. Changun Lee- Undertook kernel porting on a Jetson TK1 board by analyzing shell scripts, kernel level assembly and C code.- Built an autonomous vehicle by using YOLO, creating a 3D map from LiDAR, using a pure pursuit controller, and analyzing CAN dbc file to send control signals to the vehicle.- Reduced environment set up time from ~5 days to 5 hours by writing several shell scripts.- Utilized: Shell (Bash) Scripting, Linux, C, Python, ROS, Git, Vim



Seoul National University

• South Korea
• Education Management
• 200 - 300 Employee

• Research Assistant

  • May 2018 - Aug 2018

  Autonomous vehicle, 2D SLAMReal-time UBIquitous System Laboratory (RUBIS)Supervised by Prof. Changun Lee- Undertook a study on fundamentals of autonomous vehicles.- Created a 2D map from LiDAR data using SLAM and improved the map quality by ~40% by incorporating dead reckoning with IMU.- Utilized: Python, ROS (Robot Operating System), SLAM, PID control, Linux, Git Autonomous vehicle, 2D SLAMReal-time UBIquitous System Laboratory (RUBIS)Supervised by Prof. Changun Lee- Undertook a study on fundamentals of autonomous vehicles.- Created a 2D map from LiDAR data using SLAM and improved the map quality by ~40% by incorporating dead reckoning with IMU.- Utilized: Python, ROS (Robot Operating System), SLAM, PID control, Linux, Git