Available for AI/Product Engineering Roles

Jaeseok Song

Industrial AI & High-Performance Systems Engineer

98%+

Vision Defect Accuracy

90%

Translation Time Reduced

0%

Order Error Rate

Ansan, South Korea
dduldduck@gmail.com

I build production AI systems that improve real workflows — from industrial vision and LLM automation to backend systems and data pipelines. I focus on measurable outcomes, reliability, and clear product impact.

GitHubContactDownload Resume (EN)이력서 다운로드 (KR)

Deployment History

Software Engineer, IT Department

KYOCERA Connector Products Korea
Sep 2022 – Present

Leading Smart Factory innovation through AI, IIoT, and Digital Twin technologies.

  • Architected a multi-vendor IIoT pipeline using Node-RED to ingest real-time telemetry from KEYENCE, Panasonic, and Mitsubishi PLCs via MC Protocol.
  • Engineered a high-velocity data ingestion layer with QuestDB (ILP Protocol), managing millions of time-series data points for real-time process monitoring.
  • Developed an industrial connector defect classification system using PyTorch and CNN architectures, achieving 95%+ accuracy in identifying micro-anomalies.
  • Optimized edge inference latency by implementing gRPC-based communication between vision sensors and analysis servers, ensuring zero-bottleneck production cycles.
  • Built a real-time warehouse visualization platform using Three.js and React, transforming legacy table-based stock data into a 3D interactive twin.
  • Increased picking precision by 40% through the implementation of a refined 4-tier XYZ coordinate system (A1-2-1-2).
  • Visualized real-time inventory dynamics through an integrated dashboard, reducing stock audit duration by 50% for high-turnover SKUs.
  • Developed a full-stack Web SCM portal (React/Spring Boot) to automate the order-to-delivery lifecycle, eliminating 90% of manual communication errors.
  • Architected a middleware synchronization layer to bridge the modern SCM platform with the legacy IBM i (AS/400) DB2 backend.
  • Led the IT infrastructure alignment for Hyundai-Kia Supplier Quality (SQ) Certification, ensuring 100% compliance in data traceability and security.
  • Automated the calculation and visualization of Process Capability Index (Cpk) using Python/Flask, reducing monthly reporting time from 5 hours to 30 seconds.
  • Digitized the entire manufacturing flow through a custom QR/Barcode ERP module, resulting in 99.9% data accuracy and ~$40,000 annual cost saving.

Knowledge Modules

2024 – 2026 (Expected)

Hanyang University

Master of Science in Manufacturing AI

Thesis Subject

"Developing a CNN-based model for detecting atypical connector defects from visual inspection data."

2016 – 2022

Sangmyung University

Bachelor of Science in Industrial & Management Engineering

Thesis Subject

"Analysis of industrial accident data to classify occupational safety risk grades."

Case Studies (Problem → Approach → Impact)

Representative projects selected for hiring review: each card shows the business problem, technical approach, and measurable outcomes.

Manufacturing AI Vision Inspection

Project Lead · Data Scientist & AI Engineer

PyTorchResNetgRPCMLOps

Problem

Defect classification quality and consistency were limited by class imbalance and noisy production-floor image data.

Approach

  • Designed an end-to-end vision pipeline from data collection to model deployment.
  • Applied ResNet-based training with weighted sampling and domain-specific augmentation.
  • Integrated edge-to-server inference with gRPC for low-latency production usage.

Impact

  • Achieved 98%+ defect classification accuracy.
  • Improved fine-grained defect typing reliability in real operations.
  • Published M.S. thesis based on production-grade outcomes.

LLM-Powered Document Translation Automation

AI Product Engineer

LLM APIAutomationDesktop AppWorkflow Design

Problem

Manual KR↔JP document translation consumed ~4 hours/day and was difficult to scale across recurring business workflows.

Approach

  • Built a desktop app integrating LLM APIs for document translation workflows.
  • Added formatting-preserving post-processing to keep layout fidelity.
  • Designed a practical human-in-the-loop review flow for operational confidence.

Impact

  • Reduced translation turnaround time by 90% (4h/day → ~30m/day).
  • Maintained high formatting fidelity for business documents.
  • Turned repetitive manual work into a repeatable AI-assisted process.

SCM Order-to-Purchase Automation

Full-Stack Engineer

Spring BootJavaREST APIJPA

Problem

Order handling relied on fragmented manual processes, increasing communication overhead and order mistakes.

Approach

  • Designed Spring Boot-based architecture for order-to-purchase workflow automation.
  • Connected modern services with legacy enterprise systems through integration layers.
  • Built operational dashboards and validation rules to reduce human error points.

Impact

  • Reduced order error rate to 0%.
  • Saved 80+ work-hours per month in operations.
  • Improved process traceability and day-to-day execution reliability.

Other Selected Works

Every PDF

Comprehensive, cross-platform PDF toolkit designed for high-fidelity document editing, merging, and encryption.

pingora-proxy-manager

High-performance, zero-downtime reverse proxy manager built on Cloudflare's Pingora. Supports Wildcard SSL & TCP/UDP Streams.

Pose Nudge

AI-powered desktop app for real-time posture correction. Detects forward head posture via webcam with minimal resource usage.

whisper_ggml_plus

High-performance Flutter plugin for Whisper AI inference using GGML.

Distributed Banking

Robust account transfer system designed for distributed environments, ensuring data consistency and reliability.

Pixelo : AI Image Upscaler

Offline AI Photo Enhancer that runs the latest AI models directly on your device, transforming low-resolution images into high-resolution masterpieces.

System Diagnostics // Tech Stack

Languages

Java
Python
JavaScript
SQL
Rust

Frameworks

Spring
React
Next.js
Flutter
Three.js
Electron
Tauri
Flask

AI / ML

PyTorch
TensorFlow
YOLOv11
MLflow
CNN

Infrastructure

Docker
Nginx
Ubuntu Server
Synology NAS
CI/CD

Databases

IBM DB2 (AS/400)
SQL
JPA
QuestDB