Jaewon Shim

Education

University of California Berkeley
Data Science B.A. | GPA: 3.9 / 4.0
Dec 2025 | Berkeley, CA

Designed and deployed an ensemble machine learning model (LightGBM, Random Forest) on 500K+ rows of laser product inspection data, achieving 92% prediction accuracy and reducing false positives by 35%.
Operated the model for real-time defect risk scoring, contributing to a 12% increase in first-pass yield and $750K annual scrap reduction.
Led migration of 20+ dashboards from Tableau to Power BI, optimizing DAX data models and ETL pipelines to cut data refresh times by 45% and improve stakeholder usability.
Automated reporting workflows using Python and Excel VBA, increasing operational efficiency by 30%, and supported quality initiatives with exploratory data analysis that reduced warranty incidents by 23%.

Facilitated online programming and statistics tutoring sessions, offering guidance to over 100 students.
Earned an Honorable IT Tutor Certificate, awarded to top 1% of tutors in the company, for raising student grades from C or below to A in 96% of tutoring sessions.

Python: Pandas, Matplotlib, Seaborn, Scikit-learn, Tensorflow, Keras, LightGBM, PyTorch, Regex, API
Database Management: SQL, Query Optimization, ETL Process, Datamart, SAP HANA, Snowflake, Smartsheet
Machine Learning: Supervised/Unsupervised Learning, Ensemble Method, Evaluation, Pipeline Automation
Deep Learning: Python Frameworks, CNN (Image Processing), RNN (Time-Series), Transfer Learning
Mathematics / Statistics: Linear Algebra, Probability, Hypothesis Testing, Regression Analysis, A/B Testing
Business Intelligence: Excel, Power BI, Tableau, Smartsheet

Developed a LightGBM-based ensemble model to predict final-stage laser product defects using over 500K inspection records, achieving 85% recall and minimizing late-stage failures.
Automated SMOTE-based resampling and feature generation pipelines in Python, reducing preprocessing time by 40% and ensuring consistent model performance over multiple retraining cycles.
Integrated predictions into Power BI to visualize risk trends across product lines and inspection stages, enabling proactive quality control and reducing unexpected scrap events by 20%.

Utilized Power BI to pinpoint issues in manufacturing or design by analyzing trends in OBQ, AFR, and WIRR.
Reduced warranty incidents by 23% in 6 months through quality benchmarking and servicing.
Allowed data-driven decision-making by aligning OBQ insights with customer reviews to prioritize areas of improvements.

Featured the implementation of LSTM and GRU architectures in deep learning modeling using the TensorFlow framework, achieving an R-squared value of 0.95 with the GRU model.
The model provides actionable insights for stock investors, aiding in optimizing their investment plans.
Forecasted Samsung stock prices for the following 10 days, advising against investment due to predicted price decline.

Performed exploratory data analysis and random forest regression modeling to predict house prices.
The project demonstrates proficiency in regression algorithms, data preprocessing, model evaluation, and hyperparameter tuning.
The final model explains 80% of the variance in house prices and the model successfully predicted the price of the target house.

Employed PostgreSQL database and advanced SQL queries to perform multivariate analysis.
Explored most infectious countries along with their corresponding death rates.
Calculated the global correlation coefficient of -0.751 between GDP and infection rate, highlighting a strong negative association and emphasizing the influence of GDP per capita on the spread of the pandemic.

London bike rides dataset was used to create moving average visualization with three customizable parameters.
Implemented a heatmap with two bar charts in the tooltip, displaying ride length and weather distribution.