윤상덕

Role: Host Application Development

• Developed and released the VQM host application in C++
• Implemented PSNR, SSIM, and VMAF comparison between bitstreams

Role: Encoder Feature Development

• Implemented keyframe insertion in the reference C encoder
• Implemented long-term reference (LTR) frame support
• Modified core encoder control logic to support the above features

Role: Host Application Development / HW-SW Co-design / HW Bring-up

• Designed an NVMe-style command/completion queue architecture that defined the HW interface requirements that led the HW team's RTL changes (HW/SW co-design), and separated a dedicated queue manager module, replacing the legacy blocking (synchronous) flow with an asynchronous design
• Designed and implemented multi-channel operation on top of the asynchronous queue architecture
• Designed and implemented a Hardware Abstraction Layer (HAL) to decouple encoder control logic from hardware
• Introduced QEMU device-model-based verification, enabling driver and e2e tests without FPGA hardware

Role: Architecture Design

• Analyzed V1 limitations (single shared queue with head-of-line blocking, excessive per-frame MMIO writes, SW-mutex-based multi-instance ownership) and proposed a V2 architecture to address each
• Proposed a control/data plane split (NVMe-style admin/data queues) to isolate slow session/control commands from the fast encode path
• Proposed a Batch List submit primitive to reduce per-frame MMIO writes
• Proposed SR-IOV HW-enforced queue ownership to replace SW-mutex-based multi-instance management

Role: Founder / Lead

• Founded and currently lead a company-wide engineering study crew as a long-term program
• Strengthened practical engineering skills across the org through study of essential programming references and core concepts

Role: Lecturer

• Delivered the company's first seminar on Claude Code
• Led the adoption of AI / productivity tooling beyond personal use, educating and spreading practices across the organization

Role: SDK refactoring

• Refactored SDK to enhance code maintainability and modularity
• Documented SDK structure and usage for seamless onboarding
• Ported NVMe driver on RTOS
• Implemented MU print via direct memory access (bypassing MBOX)
• Built a GoogleTest-style test framework and MU kernel test cases

Role: Hardware IP Driver Development

• Developed and verified third-party IP drivers for I2C(SMBUS), UART, and SPI
• Developed and verified XCENA IP drivers for GMON, RAU, GDMA, CMDS, MBOX, and MU
• Established driver verification framework leveraging Zebu and HAPS platform
• Collaborated with the SoC team for IP verification

Role: ROM/RAM code develop and SDK porting

• Developed an automated deployment script for Zebu and HAPS platform
• Developed ROM/RAM code on ARM
• Ported SDK on ARM

Role: Lecturer

• Taught How to Co-Work using Coding Style and Git
• Implemented and taught state-of-the-art architecture

Role: AI Part Lead / Deep Learning Development

• Wrote an R&D proposal
• Developed a deep-learning model for sea-level observation
• Researched Video Enhancement using Deep Learning
• Developed Tsunami Detection Algorithm with Anomaly Detection

Role: Feature Improvement and Build Cluster with MPI

• Developed Water Gate and Wheel Module for Hydrodynamics Simulator using MPI with Fortran
• Built a Cluster with MPI

Role: System Engineer

• Improved performance by refactoring legacy code
• Built and Deployed Deep Learning Application with Docker

Role: Develop VPN Modules

• Developed Thread-safe Runtime for VPN with C/C++
• Developed Microservice Controller for Security Apps (FW, VPN, IDS, Anti-Virus) with Docker
• Researched and Developed Network Anomaly Detection using New Features with TensorFlow
• Developed Kernel Module of Firewall

Role: Maintain Web UI

• Maintained Web UI of the VPN
• Developed New Features

Role: Develop Web UI

• Visualized cyber threat intelligence

Role: Research Assistant

• Developed a Cloud Resource Management System on OpenStack and Eucalyptus
• Developed a Web UI with JSP
• Developed a Cloud Resource Schedule

Role: Research Assistant

• Developed a Cloud Resource Management System
• Deployed distributed frameworks (MPI, MapReduce) on virtualized environments
• Developed Volume Rendering with CUDA

Role: Framework Design and Develop

• Wrote an R&D proposal
• Developed Runtime System for Deep Learning Inference based on GPU, Xeon Phi using Docker

Role: Team Leader

• Wrote an R&D proposal
• Optimized Deep Learning Training with GPU Profiling
• Reduced GPU idle time through profiling
• Tuned GPU memory layout for compute