Joshua Chung Email & Phone Number

Remote, Us

Boise, Idaho, Us

- Developed a clustering model using HDBSCAN and UMAP to classify stores into different investmentstrategies, guiding $600 million in national investment to lower prices and capture market share.Created an interactive map visualization for business stakeholders to explore model-based investmentrecommendations.- Employed NLP techniques, PySpark, and open-source LLMs to automate text cleaning and attribution of 5,000,000 UPCs, resulting in a master item dataset foundational for future initiatives basedon similar item attributes and clean item descriptions, leading to an estimated productivity savingsof $60 million.- Currently optimizing shelf planograms and stock order sizing for different items to maximize profitability and minimize out-of-stock events based on sales velocity and customer preferences per store.

West Jordan, Utah, Us

- Spearheaded an effort to increase profit margins on multiple products by 50-100% by creating new product KPIs which identify market opportunities using NLP deep learning techniques and unsupervised learning methods on millions of product SKUs and then using said market opportunities to A/B test new prices.- Improved dropship sales time series modeling performance (r-squared score) by 20% by engineering 250+ new features using Python and SQL along with utilizing computationally expensive modeling techniques only feasible by using cutting-edge cloud GPU software and distributed computation backends.- Widened the scope of possible data science projects by decreasing ML model training times by over 50% by scaling up commonly used algorithms through Dask, a distributed computation software, Nvidia RAPIDS, a GPU backend ML package, and AWS cloud computing.- Consistently worked cross-functionally with Marketing, Growth, Sales, and C-Suite executives by collaborating on designing data science projects that drive value and presenting these data science applications in a low-technical manner.

Berkeley, Ca, Us

- Wrote paper and software on new data science framework for data science competitions, using sensitivedata. Drafted all code necessary for analysis in paper, as well as custom modeling functions, to utilizeHPC environment and NERSC supercomputer.- Developed innovative approach for data-driven dimensionality reduction on Numerai’s heavily obfuscated data set of 500,000 observations. Fostered improvements in model metrics to allow public datascience competitions to be feasible in domains with sensitive but informative data.- Sped up run time by 40- to 60-fold for common Scikit-learn algorithms by engineering software,utilizing massively distributed programming via high-performance computing (algorithms includedhyperparameter cross-validation, feature selection, and computing non-Euclidean distance matricesfor feature clustering).- Improved Random Forest performance from benchmark .03 Spearman Rank Correlation to competitive .05 by tuning and testing 20 combinations of dimensionality reduction techniques for featureselection and extraction, and utilizing modern hyperparameter tuning techniques.

Berkeley, Ca, Us

- Achieved an AUC-ROC of .91 for daily binary forecasts of Apple stock prices by using 500,000 rows of training data, TensorFlow libraries, commonly used neural network architectures, and LSTMs. - Reduced the Apple stock dataset size by 40x using a Pandas ETL pipeline to transform the 21 million transactional data points into daily time steps to fit into Google Colab memory and to engineer 10 other informative features.- Utilized incremental learning and walk-forward validation with time-series data to simulate the potential profit outcome of using the neural network in the stock market with a data stream.- Reduced the Apple stock dataset size by 40x using an ETL pipeline through Pandas chunking to transform the 21 million transactional data points into discrete bins to load into Google Colab memory and to engineer 10 other informative features.- Utilized incremental learning and walk-forward validation to avoid p-hacking with time-series data and to simulate a real-world scenario of using the neural network in the stock market with a data stream.

Berkeley, California, Us

- Collaborated with an undergraduate research team to build software utilizing Python, Pandas, and Numpy for the Public Editor fake news detection platform, where users can collectively rank the credibility of news articles on the web.- Designed inter-annotator agreement algorithms that work around low data depth and skewed distributions from users’ inputs.- Developed a custom multiplicative weighting algorithm to rank user reputation for gamification and model optimization.

Berkeley, Ca, Us

-Direct trainings for staff, emphasizing efficient communication and professionalism.-Supervise staff during shifts and meetings to meet security standards for dormitories.-Coordinate scheduling for staff and track employee performance to fairly delegate responsibilities.

Berkeley, Ca, Us

-Checked in students at UC dormitories, providing friendly and comforting environment to students. -Maintained the safety of residents through comprehensive security check-ins and patrols.-Trained in CPR, pepper spray, and active shooter protocol.

Master'S Degree, Information And Data Science

Bachelor’S Degree, Data Science

High School