Seungho Jang

Seungho Jang

Graphics Engineer & Computer Vision Researcher

PDF
Location
Seoul, Republic of Korea, South Korea
Email
Phone
(+82) 10-5610-6541
Website
https://sjang1594.github.io/
GitHub
sjang1594
LinkedIn
Seungho Jang
YouTube
nugabab_life

Experience

– present

Platform Engineer (Unreal) at Hyundai AutoEver

Engineered real-time factory monitoring and digital twin systems integrating Unreal Engine with NVIDIA Isaac Sim across HMG manufacturing sites.

Highlights

  • Built real-time factory line monitoring system on NVIDIA Isaac Sim with TCP-based robotic arm control via Lula IK — enabling sub-100ms remote actuation for KIA production R&D.
  • Architecting UE5-based real-time 3D industrial twins for shipyard, steel plant, and automotive assembly operations across 3 HMG group sites.
  • Designed IoT data pipeline (Valkey/Redis → Unreal runtime) streaming live asset state to operator dashboards at 30+ Hz.
  • Implementing replay system for incident root-cause analysis and WebRTC remote streaming for distributed monitoring.
  • Built OpenFOAM → OpenVDB conversion pipeline and UE5.3 VDB Sequencer plugin for Hyundai Steel CFD digital twin — per-tick PPM computation and VDB field cutoff streamed to operator dashboard.
  • Optimized POD-ROM reconstruction solver (C++17, Eigen, OpenMP parallel for simd) — POD basis caching cut per-input latency from 107ms → 23ms (4.6×) and disk I/O from 41 iterations to 1 (12.5×) on 240-core MPI cluster.

–

Independent Technical Development at Self-Directed

Self-directed deep-dive into GPU rendering systems and CUDA kernel development, shipping open-source work targeting graphics and DL infrastructure roles.

Highlights

  • Built Luna Engine — backend-agnostic real-time renderer with Vulkan and DirectX 12 backends, built-in GPU profiler, and Isaac Sim interop layer.
  • Implemented Flash Attention forward and backward kernels in CUDA — tiling, online softmax, shared-memory bandwidth optimization on NVIDIA Ampere/Hopper.
  • Deep-dive study of DirectX 11 graphics pipeline and modern algorithm design (data structures, graph algorithms, dynamic programming).

–

Sensor Software Engineer at MORAI Inc

Shipped production sensor simulation software for autonomous driving — owning virtual sensor plugins, scenario execution, and labeling pipelines used in AV development workflows.

Highlights

  • Designed and shipped USF shader-based radar simulation in Unreal Engine modeling chirp count, sample rate, RX antenna geometry, and antenna pattern; produced range-Doppler maps via Python/ROS pipeline. First shader-native radar in MORAI's product line. [Patent WO2024/117564]
  • Architected scenario runner from scratch implementing ASAM OpenSCENARIO — load/edit/save .xosc, gRPC simulator integration, OBB/SAT collision detection, batch simulation Python API. Adopted as MORAI's primary scenario execution engine.
  • Built bounding box labeler supporting vehicles (part segmentation), pedestrians (animation states), and obstacles — 4 coordinate frames (Camera/LiDAR/Vehicle/ENU), dual representation (8-corner / center-extent).
  • Authored UE5 Coordinate Converter plugin for bidirectional left/right-handed transforms (NED, ENU, AER) with unit tests — eliminated cross-team integration errors.
  • Delivered LiDAR motion distortion post-processing via SLERP rotation interpolation and coordinate transformation, integrated into MORAI's LiDAR sensor pipeline.

–

Graduate Teaching Assistant at University of Missouri - Saint Louis

Designed and ran C++ labs for OOP, memory management, STL, and graph algorithms across three semesters.

–

Teaching Assistant at Washington University in Saint Louis

Supplemental instruction for Signals & Systems and Engineering Mathematics.

Education

–

Master in Computer Science
  from University of Missouri - Saint Louis with GPA of 4.0 / 4.0

Courses

  • Operating System
  • Computer System
  • Deep Learning & Machine Learning
  • Image Processing & Computer Vision
  • Artificial Intelligence (CS4130/5130)
  • Advanced Algorithm

–

Bachelor in Electrical & Electronic Engineering
  from Washington University in Saint Louis with GPA of 3.8 / 4.0

Courses

  • Engineering Mathematics
  • Signals & Systems
  • Radar Systems

Graduate Certificate in Artificial Intelligence — UMSL / NSA / DHS National Center

Robotic Software Engineering — Udacity

Computer Vision — Udacity

Introduction to Computer Graphics with DirectX 11 — Part 2 Realtime Pipeline

Projects

Luna Engine :

– present

Backend-agnostic real-time rendering engine with Vulkan and DirectX 12 backends. Features built-in GPU profiler for frame timing and pipeline-stage attribution, and an Isaac Sim interop layer for streaming simulation state into the renderer.

Highlights

  • C++ / Vulkan / DirectX 12
  • GPU Profiler
  • HAL Design Pattern
  • Isaac Sim Interop
  • Premake / ImGui

Flash Attention CUDA Kernel :

– present

From-scratch CUDA implementation of Flash Attention forward and backward passes with tiling and online softmax. Studied shared memory and HBM bandwidth tradeoffs on Ampere/Hopper architectures.

Highlights

  • CUDA
  • Tiled GEMM
  • Online Softmax
  • Memory-efficient Attention

GPU Gaussian Splatting Rasterizer :

– present

Custom CUDA rasterizer for 3D Gaussian Splatting implementing forward pass tile-based rasterization, alpha compositing, and gradient flow through Gaussian parameters (mean, covariance, opacity, spherical harmonics).

Highlights

  • CUDA
  • 3D Gaussian Splatting
  • Alpha Compositing
  • Neural Rendering

Hyundai Steel CFD Digital Twin (POD-ROM) :

– present

Real-time CFD visualization pipeline for Hyundai Steel digital twin: converted OpenFOAM simulation output to OpenVDB for UE5.3 streaming, built a custom VDB Sequencer Unreal plugin for per-tick PPM computation and field cutoff, and optimized the POD-ROM reconstruction solver — reducing per-input latency 4.6× and disk I/O 12.5× on a 240-core MPI cluster.

Highlights

  • OpenFOAM / OpenVDB / UE5.3
  • POD-ROM Solver
  • C++17 / Eigen / OpenMP SIMD
  • VDB Sequencer Plugin
  • CFD / Digital Twin

Hyundai UAM — Radar Point Cloud Visualization (UE5) :

–

Gaussian random distribution model for realistic radar returns with raycast-based aerial-only target isolation. Improved point-cloud fidelity for UAM perception R&D.

Highlights

  • Unreal Engine 5
  • Radar Simulation
  • Point Cloud

Project NIA — AV Edge-Case Dataset :

–

Authored, modified, and validated 1,200 OpenSCENARIO edge-case scenarios across 6 Korean HD maps for the Korean MOLIT autonomous driving dataset.

Highlights

  • ASAM OpenSCENARIO
  • HDMap
  • Scenario Authoring

Samsung Synthetic Data Generation :

–

Authored complex multi-agent scenarios (15+ dynamic objects per batch) using DataGen, with detailed semantic annotation for perception model training.

Highlights

  • Synthetic Data
  • Scenario Authoring
  • Semantic Annotation

Patents

WO2024/117564 — Scenario-Based Autonomous Driving Vehicle Simulation Method and System

- Author: Eungback Kim, Seungho Jang, Seongyeon Park, Hoseup Lee, Hein Jo

- Published Date: June 2024

KR 10-2024-0076717 — Electronic Device and Method for Processing Point Cloud Data

- Author: Heecheol Yoo, Seungho Jang, Hojun Lim

- Published Date: June 2024

Volunteer

3D Point Cloud Data Processing

- Presenter at Spatial AI KR

- Link

Understanding and Implementing SLAM with NVIDIA Jetson Nano

- Leader at MODULABS

Brain Dynamics and Control Research Group

- Undergraduate Researcher at Washington University in Saint Louis

Computing Club - Camera Control System

- Member at University of Missouri - Saint Louis

Awards

Sweeney Memorial Scholarship
  from Washington University in Saint Louis — McKelvey School of Engineering

Robert Heider Engineering Scholarship
  from Washington University in Saint Louis — McKelvey School of Engineering

Languages

Korean
Fluency: Native Speaker
English
Fluency: Native Speaker

Skills

C++
Level: Master
Keywords:
  • Modern C++ (17/20)
  • System Concurrent Programming
  • Rendering Engine Architecture
  • HAL Design Pattern
CUDA / GPU Compute
Level: Advanced
Keywords:
  • Custom Kernel Development
  • Tiled GEMM
  • Memory Hierarchy Optimization
  • Flash Attention
  • Gaussian Splatting Rasterizer
Graphics APIs
Level: Master
Keywords:
  • Vulkan
  • DirectX 12
  • DirectX 11
  • HLSL / USF
  • Render Graph
  • PBR / Ray Tracing
Unreal Engine
Level: Master
Keywords:
  • Unreal Engine 4 / 5
  • Shader Programming (.usf)
  • Virtual Sensor Plugin
  • Coordinate Converter Plugin
  • Digital Twin / Factory Sim
  • RenderDoc
Simulation Platforms
Level: Master
Keywords:
  • NVIDIA Isaac Sim
  • NVIDIA Omniverse
  • Unity
  • ROS
  • ASAM OpenSCENARIO
Computer Vision
Level: Advanced
Keywords:
  • Multi-view Geometry
  • SfM / Bundle Adjustment
  • 3D Reconstruction
  • Point Cloud Processing
  • OpenCV / Open3D
ML / Neural Rendering
Level: Advanced
Keywords:
  • PyTorch
  • TensorFlow
  • 3D Gaussian Splatting
  • NeRF
  • Diffusion Models
Python
Level: Master
Keywords:
  • NumPy / SciPy
  • PyQt
  • PyTest
  • gRPC
  • ROS
Languages
Level: Intermediate
Keywords:
  • C
  • C#
  • Java
  • JavaScript

Interests

Activity
Keywords:
  • Soccer
  • Swimming
Gaming
Keywords:
  • Starcraft
  • Overwatch2
  • FIFA Online