Predicting Confirmed Cases of Various Epidemics Using Global Temporal-Feature-Based Graph Convolutional Network
Proposed GLOGCN, a graph convolutional network that leverages global temporal patterns for epidemic case forecasting.
View paperJisu Kang
AI/ML Engineer
I build agentic AI systems, retrieval workflows, and applied machine learning products with Python, TypeScript, and graph-powered context.
About
A bit about myself.
I'm an AI/ML engineer at LG Electronics, focused on practical AI agents for enterprise workflows and manufacturing environments.
My recent work spans NL2SQL evaluation automation, document-based RAG, VLM-powered safety detection, and a local AI assistant platform with installable plugins.
Publications
Peer-reviewed research.
Beyond Virtual Points: Depth-Enhanced LiDAR-only 3D Object Detection with Semi-Supervised Learning
Developed a semi-supervised LiDAR-only 3D object detection framework that improves the speed-accuracy trade-off.
View paperLeveraging Out-of-Distribution Unlabeled Images: Semi-Supervised Semantic Segmentation with an Open-Vocabulary Model
Used open-vocabulary segmentation to generate reliable pseudo-labels for semi-supervised semantic segmentation.
View paper2D Hole-Arrayed Double-Anode Structure Exciting Surface Plasmon Polaritons for Enhancing Outcoupling Efficiency of Organic Light-Emitting Diodes on Silicon Wafers
Designed a double-anode OLED structure to enhance outcoupling efficiency by controlling surface plasmon polariton behavior.
View paperTemporal-Contextual Attention Network for Solid-State Drive Failure Prediction in Data Centers
Introduced TCAN, an LSTM-Transformer-based attention network for SSD failure prediction.
View paperProjects
Selected AI work and research.
NL2SQL Evaluation Agent2025
An end-to-end evaluator that scores and improves internally developed NL2SQL agents with feedback-driven automation.
Graph-RAG Chat Agent2025
A document-based internal knowledge assistant that uses graph database context to improve retrieval quality.
CCTV Detection Agent2026
A customizable manufacturing safety detection agent for events such as fire, fallen persons, and missing masks.
Local AI Assistant Platform2026
An Electron-based local assistant platform with installable plugins for personalized work automation.
UX Guideline Agent2025
A hackathon-winning AI agent that analyzes UX guidelines and application screens to detect violations.
Work Experience
My career so far.
Mar 2025 - Present
AI/ML Engineer · LG Electronics
Seoul, South Korea
Developing internal AI agents and local assistant platforms that connect LLMs, VLMs, RAG, graph databases, and production workflows.
- Built an NL2SQL evaluation automation agent with feedback-driven scoring, improving internal benchmark performance by about 10%.
- Enhanced a document-based RAG chat agent with graph database context, improving internal benchmark performance by about 15%.
- Developing a customizable CCTV detection agent for manufacturing safety and compliance events, preparing for deployment across about 3,000 cameras.
- Developing an Electron-based local AI assistant platform with plugins for meeting summarization and local file search or organization.
Skills
Core tools and systems.
Languages
AI/ML Frameworks
Tools & Databases
Education
Academic background.
Korea University
Seoul, South Korea
Master of Engineering in Industrial and Management EngineeringGPA 4.20/4.50
Seoul Women's University
Seoul, South Korea
Bachelor of Business AdministrationGPA 4.45/4.50
Bachelor of Engineering in Software ConvergenceGPA 4.15/4.50
Awards
Recognition and certificates.
1st Place, LGE Hackathon
2025Developed an AI agent program that analyzes UX guidelines and in-progress application screens to identify guideline violations using RAG, LangGraph, VLMs, and LLM agents.
Blog
Project deep-dives and engineering notes.
Building a Customizable CCTV Safety Detection Agent2026-03
Improving Enterprise RAG with Graph Database Context2025-12
Automating NL2SQL Evaluation with Feedback-Driven Scoring2025-09
UX Guideline Violation Detection with LangGraph and VLMs2025-06
Contact
Email and professional links.