Yeeun Jeon

Yeeun Jeon

AI Engineer

I'm an AI engineer with a Data Science background. My focus is RAG and LLM Agents — and what interests me most is solving problems inside real industrial domains. Right now I'm working my way toward being a full-stack engineer — so that I can own an AI product end to end, from model to interface.

AI / LLM
LangChain LLM APIs PyTorch
LANGUAGES
Python TypeScript JavaScript
FRONTEND & BACKEND
React Node.js FastAPI
DATA & INFRA
PostgreSQL Docker Git / GitHub

Experience.

Samsung AMR
  • Proposed an algorithm for computing optimal standby positions for AMRs (Autonomous Mobile Robots), reducing unloading transport time and total transfer time
  • Analyzed OHT (Overhead Hoist Transport) data and predicted section traversal time — a Temporal Fusion Transformer beat the baseline on MAE
★ Ranked 1st of 8 interns in overall evaluation — selected as Top Intern

  • Built a RAG-based chatbot for insurance planners
  • Implemented the whole pipeline end to end: document parsing → chunking → retrieval → LLM generation → RAG evaluation
  • Tuned it for the auto-insurance document domain with query transformation, ensemble retrieval, and prompt engineering
Naïve RAG62.3%
Final Architecture96.6%
Hanwha RAG

Samsung Query Rewrite
  • Implemented the Query Rewrite module inside DS Assistant — generates two query types at once from the prior conversation and the latest question
  • Retrieval Query (keyword-oriented) + Generation Query (natural sentence form)
  • Built the pipeline: query → document reranking → answer generation driven by the Generation Query
  • Optimized for perceived response speed by using an LLM with low TTFT (Time to First Token)
  • Built an evaluation set from real user data; measured Recall and nDCG against the number of retrieved / reranked documents

  • Retrieves recent articles on AI initiatives across other financial firms
  • LLM-based summarization and insight extraction from an insurer's point of view
  • Formats everything into the internal report template — built as an end-to-end pipeline
Related code → github.com/esthyj/ai-weekly-report
Hanwha Insurance Report Agent

Publications.

IEEE · 2024

Digital Health Sensor Data in Autism: Developing Few-Shot Learning Approaches for Traditional Machine Learning Classifiers

ieeexplore.ieee.org/document/10780502 →

SMOTE_FSL — a data augmentation algorithm that lets traditional ML models such as Random Forest and SVM train effectively even on very small datasets. Standard SMOTE only synthesizes samples inside the minority class distribution, but when the sample is small — as clinical data usually is — the collected data rarely represents the full characteristics of a disease. SMOTE_FSL applies Statistical Inversion to the padding parameter, deliberately allowing synthetic samples to be generated beyond the range of the original data.

Education.

Hanyang University

B.S. in Data Science · Mar 2021 – Aug 2025 SUMMA CUM LAUDE
Probability & Statistics Data Structures Machine Learning I · II Algorithms & Problem Solving Intelligent Robotics Human-Computer Interaction Artificial Intelligence Data Science Computer Vision
Hanyang Data Science

Awards & Certifications.

AWARD

Presidential Science Scholar

Korea Student Aid Foundation · Feb 2024
CERTIFICATIONS
CertificationLevelValid
SQLDPassedJun 2024 –
AdSPPassedSep 2024 –
OPIcIntermediate HighFeb 2026 – Feb 2028
Presidential Science Scholarship award ceremony

Let's Connect!

Always open to new collaborations and conversations.

EMAIL esthyj@naver.com
PHONE +82 10-2468-7535
GITHUB github.com/esthyj
LINKEDIN yeeun-jeon-belu