Data-driven Operations
I build the pipelines and dashboards that operations teams trust.
Data Engineer
- ETL Pipelines
- Data Modeling
- BI Reporting
- Workflow Automation
About

I grew up in South Korea and moved to the Netherlands in December 2023, after an exchange semester convinced me to build a career somewhere new. With an MSc in Data Science and Society from Tilburg University, what draws me to data work is simple: I love digging deep into one problem and proving the answer with numbers.
Day to day, I build dbt and SQL pipelines, KPI dashboards, and internal platforms that replace manual reporting and tracking. At Reederei Nord, automating 20+ KPIs cut reporting and verification effort by over 90%. Lately I have been exploring how AI can take over the repetitive parts of a colleague's job so they can focus on what needs a human, from RAG-based chatbots and prompt engineering to token efficiency, and I am building my own AI-powered apps on the side.
Outside of work I like slow weekends just as much as I like being somewhere new. Travel is my reset button.
Skills
Languages
- Python
- SQL
Databases
- PostgreSQL
- MySQL
- MS SQL Server
Data Engineering
- dbt
- Git
BI & Reporting
- Excel
- Power BI
- Streamlit
Workflow & Docs
- Markdown
- YAML
AI
- GenAI & LLM APIs
- Prompt engineering
- RAG chatbots
- LLM-assisted development
Currently exploring
Career

Reederei Nord Group
2026.08 - PresentData Engineer - Amsterdam, Netherlands
- Extracted and joined purchase data from 4 fragmented catalogue systems via primary keys, grouping inconsistent item-level categories into standardized groupings through an automated dbt transform-validate-load pipeline. Deployed the resulting dashboard (Streamlit) via a cloud server and application deployment platform, enabling simultaneous access across multiple regional branches. Unified 7,000+ annual purchase orders into ship-, year-, vendor-, and port-level views, shifting vendor negotiations from experience-based judgment to data-backed leverage.
- Designed and automated dbt pipelines to calculate and refresh 20+ operational KPIs live directly from source ERP data, replacing manual, record-by-record metric verification. Restructured the review process so decision-makers assess only final KPI outputs and the specific records behind flagged exceptions, rather than the full underlying dataset. Reduced end-to-end reporting and verification effort by over 90%, reallocating analyst time from routine checks to exception handling and strategic reporting.
- Built an LLM-based audit pipeline to evaluate vessel inspection records against PSC (Port State Control) and SIRE compliance standards, using prompt-driven classification to flag deficiencies embedded in free-text ERP data. Configured the model to output its reasoning alongside each flagged result, enabling reviewers to cross-check findings rather than re-derive them from raw records. Improved audit consistency across reviewers and surfaced deficiencies that manual review had previously missed, raising the overall data quality of compliance reporting.
- Designed and launched a company-wide internal platform for submitting, reviewing, and approving AI and data analysis requests, replacing unstructured coordination via email and chat. Built end-to-end progress tracking, file management, and admin functionality, backed by a cloud server managing request and user data. Established a transparent, trackable workflow for the CI team, improving task allocation and prioritization across the analyst team.

HD Hyundai Marine Solution Europe B.V.
2025.07 - 2026.07Logistics & Data Analysis Officer - Rotterdam, Netherlands
- Redesigned a one-size-fits-all demand-planning process using ADI/CV2 classification and multi-model forecasting (Croston, ARIMA, LightGBM, XGBoost) with MASE-based evaluation, directly informing leadership replenishment and procurement strategy; separately built a SKU-level Lead Time risk framework later adopted as an internal planning standard across teams.
- Designed and owned end-to-end Python/SQL ETL pipelines to replace fragmented, manually reconciled ERP exports, consolidating 20,000+ daily ERP and logistics records into a centralized PostgreSQL warehouse and cutting reporting cycle time by roughly 80%.
- Consolidated 5 years of SKU-level pricing data and applied SQL and linear regression to model expected pricing trajectories after external auditors flagged year-over-year purchase-to-sale price gaps as a financial risk, validating the discrepancies as normal trends and achieving 100% pricing integrity verification across the European warehouse's full SKU inventory.
Republic of Korea Army
2019.12 - 2021.06Sergeant, Corps Transportation Unit - Pocheon-si, Gyeonggi-do, Republic of Korea
- Served as a driver in the Corps Transportation Unit, supporting logistics supply, operations, and training exercises, contributing to 30+ successful training and field exercises without incident.
- Promoted to squad leader, leading a squad of 10+ personnel and coordinating daily task assignments and readiness; recognized with Squad of the Month and Soldier of the Month (x2) awards.
- Responsible for peer counseling, providing mental health counseling and support to fellow soldiers to help maintain unit morale and cohesion.
- Contributed to unit-wide COVID-19 prevention and safety measures during the pandemic.
Projects

CatchCook: AI Recipe Extraction App
- Building a React Native (Expo) app for iOS and Android that turns a YouTube cooking video link into a structured recipe with ingredients, measurements, and timestamped steps.
- Uses the Gemini API to watch the video directly (including measurements shown only as on-screen text), with prompt engineering and a selective-viewing strategy to cut latency and token usage.

Predicting Crowdfunding Success: Multimodal Deep Learning
- Benchmarked Logistic Regression, MLP, and a Multimodal Transformer on 245K+ Kickstarter campaigns, combining structured features with RoBERTa embeddings (pre-trained vs fine-tuned).
- Best model (MLP with structured + text fusion) reached F1 0.869 / AUC-ROC 0.907, outperforming the Transformer and showing that added architectural complexity did not improve predictive performance in this setting.
Customer Segmentation: RFM Scoring vs. K-Means Clustering
- Merged 5 fragmented e-commerce data sources into a single behavioral view of 1,468 customers and ~53,000 transactions, engineering 8 features beyond traditional RFM and validating K-Means against Agglomerative Clustering (ARI 0.876).
- Quantified the business impact: traditional RFM quintile scoring agreed with the clusters only 49% of the time, misclassifying 335 high-value repeat buyers worth $879,941 as low-value.
Portfolio Website (Next.js)
- Built and deployed a personal portfolio site with Next.js, Tailwind CSS, and Framer Motion.
- Implemented scroll-based reveal animations and a splash-screen intro sequence.
Education

MSc Data Science and Society
Tilburg University
GPA 8 / 10 - Graduated with Distinction
Specialization: Business
Thesis: Predicting the success of crowdfunding approach by multi-layer perceptron and multimodal transformer

BA in Political Science & Public Relations (Double Major)
2017 - 2024Dankook University
GPA 90.1 / 100
Exchange: FHWien der WKW, Vienna, Austria (Feb - Aug 2022)
- Globalization Scholarship (Spring 2022)
- Special Scholarship (Spring 2022, Fall 2022)
- Academic Excellence Scholarship (Jan 2023)