SeoJin Yoon윤서진

Senior AI Researcher / 선임AI연구원

Building intelligent systems from scratch — Reinforcement Learning, Computer Vision, and Deep Learning at HD Korea Shipbuilding & Offshore Engineering.


About

A researcher who learns by building from scratch / 밑바닥부터 구현하며 배우는 연구자

I'm a Senior AI Researcher at HD Korea Shipbuilding & Offshore Engineering (AI Center), working on RL-based steel plate nesting optimization and speech recognition R&D.

HD한국조선해양 AI Center 소속 선임AI연구원으로 조선설계 네스팅 강화학습 최적화 및 음성인식 R&D를 담당하고 있습니다.

I believe the best way to understand any algorithm is to implement it from scratch. My GitHub is a collection of "from-scratch" implementations — from language models to world-model RL agents, diffusion models to autograd engines.

Reinforcement Learning Computer Vision Deep Learning LLM / NLP Autonomous Systems Speech Recognition

Experience

Career timeline / 경력 타임라인

Senior AI Researcher / 선임AI연구원
HD Korea Shipbuilding & Offshore Engineering / HD한국조선해양
Present
  • RL-based steel plate nesting optimization for ship design (조선설계 네스팅 강화학습 최적화)
  • Speech recognition R&D (음성인식 연구개발)
  • AI Center (AI센터)
Data Analyst / Engineer
Previous Roles / 이전 경력
Prior
  • SNS Commerce Data Analysis (SNS 커머스 데이터 분석)
  • Web Data Analysis Solutions (웹 데이터 분석 솔루션)
  • LG CNS Smart SMA

From-Scratch Series

Core projects — implementing algorithms from the ground up / 핵심 프로젝트 — 알고리즘을 밑바닥부터 구현

🌍 dreamer-from-scratch

DreamerV3 World Model RL agent — learning to act by imagining future trajectories in a learned world model.

Reinforcement Learning World Models DreamerV3
View on GitHub →

📝 llm-from-scratch

GPT-style language model built from scratch — tokenizer, attention, training loop, and text generation.

Deep Learning Transformers GPT
View on GitHub →

🎨 diffusion-from-scratch

DDPM and DDIM diffusion models — understanding the math and implementation of denoising diffusion.

Generative AI DDPM DDIM
View on GitHub →

🔥 pytorch-from-scratch

Linear regression to Transformers — a progressive deep learning journey implementing every layer by hand.

PyTorch Deep Learning Education
View on GitHub →

⚙️ autograd-engine

Automatic differentiation engine — the backbone of modern deep learning, implemented from first principles.

Autograd Backpropagation Core ML
View on GitHub →

🎮 simpleRL

From DQN to PPO — a clean, educational implementation of classic reinforcement learning algorithms.

DQN PPO Reinforcement Learning
View on GitHub →

Other Projects

Applied AI and tools / 응용 AI 및 도구

🔌 mcp-python-tutorial

Comprehensive MCP (Model Context Protocol) tutorial in Python — building AI tool integrations step by step.

MCP Python Tutorial
View on GitHub →

📦 WareHouseEnv

Reinforcement learning for warehouse logistics optimization — custom Gym environment for material handling.

Reinforcement Learning Logistics OpenAI Gym
View on GitHub →

⚫ gomoku_puct

Gomoku (Five in a Row) AI using Monte Carlo Tree Search with PUCT — AlphaZero-style game playing.

MCTS Game AI AlphaZero
View on GitHub →

Blog Highlights

tildalice.io — Writing about AI, Deep Learning, and Software Engineering

765
Posts
76,992
Total Views
10+
Categories

Most Popular Posts

Data Analysis80 Humor79 Finance & Quant64 Python Tips64 Reinforcement Learning63 AI/Deep Learning61 CBM/PHM53 Coding Interview48 News & Commentary46 LLM44

Skills

Technical stack / 기술 스택

Languages

Python C JavaScript

ML / Deep Learning

PyTorch TensorFlow Scikit-learn

Reinforcement Learning

PPO SAC DQN DreamerV3

Computer Vision

YOLO ViT SAM OCR

NLP / LLM

Transformers Fine-tuning RAG

Data Engineering

Pandas Polars SQL Spark

Tools & Infrastructure

Docker Git FastAPI Streamlit
🎮

Fun fact: In high school, I built a custom game in Warcraft 3 World Editor called "호조아케이드" (HoJo Arcade). That early experience with game logic and scripting sparked my interest in programming and eventually led me to AI research.


Get in Touch

Interested in AI research, collaboration, or just want to chat? / 연락 주세요!