Derek Goh
I am a currently a senior computer science student at the University of Michigan pursuing full-time software engineering and machine learning engineering jobs. I enjoy applying my computer science abilities to solve real world problems.
Resume
Experience
Software Engineering / Machine Learning Engineering Intern, STR
I developed a machine translation module for a data extraction framework using PyTorch and NLLB-200. I containerized and deployed models with Docker and indexed the model outputs to Elasticsearch with Rest APIs. I also orchestrated and integrated models into CI/CD pipelines and collaborated in Agile Scrum meetings. I conducted research on and implemented domain adaptation techniques to optimize for technical vocabulary such as LoRA, Bilingual Lexicon Creation, Lexically-Constrained Beam Search, and Domain Adversarial Training. By using a combination of these methods on 62000 training documents, I increased the BLEU, ROUGE, and BERTScore of the translated documents by 77%, 32%, and 6% respectively.
Web Developer, AnomalyWC (2023 - Present)
I migrated the corporate website (anomalywc.com) for this electronics startup from WordPress to a custom site built with React and JavaScript, improving the user experience for 450+ customers. This resulted in the reduction of operating costs by hundreds of dollars annually.
Software Engineering / Machine Learning Engineering Intern, George Mason University ASSIP (2021)
I worked with a team to develop Python models with SafeGraph data to predict user locations for COVID-19 contact tracing. I built these models with TensorFlow and Scikit-Learn, and analyzed and visualized them with Pandas, NumPy, and Matplotlib. I then wrote a report, Spatio-Temporal Prediction of Human Mobility, which described the results. My work was conducted under the guidance of Dr. Andreas Zufle and supported by the National Science Foundation. It was presented at the GMU Aspiring Scientists Summer Internship Program Research Symposium.
Projects
BERT, Twitter API, Pandas, Matplotlib
Twitter Political Partisanship Identification
Social media misinformation can lead to political polarization and extremism. To combat this, I worked with a team to fine tune a BERT model to predict the political partisanship of Twitter tweets, achieving a 91% Accuracy and F1-Score. Our dataset was curated by scraping 10000 tweets with the Twitter API, preprocessing, and storing them to a Pandas dataframe. We conducted a grid search to optimize for hyperparameters such as learning rate and weight decay. We plotted and analyzed our results with Matplotlib and conducted error analysis to identify trends in misclassified tweets.
React Native, XCode, React-Native-ble-plx, UI Kitten, React Native Gifted Charts
Wearable Sunlight, Vitamin D, and UV Exposure Tracker
Many Midwesterners suffer health problems arising from improper sunlight exposure, especially in the winter. To remedy this, I worked with a team to develop a wearable Bluetooth Low Energy Sunlight, Vitamin D, and UV exposure LTR390 sensor and corresponding React Native mobile app that allow users to view and set their desired exposure levels. The app features a summary dashboard, exposure history charts and calendar, preferences, and desired exposure slider panes.
iOS Development, SwiftUI, AR Kit, Reality Kit
AR Garden Planner
Poor landscaping decisions such as planting trees too close to houses or power lines can cause severe property damage. To prevent this, I created an iOS AR app that checks landscape space and climate compatibility using Swift, AR Kit, and RealityKit. It lets users place life-size plant models in their yards, keeping plant sizes correct up to 25 meters and ensuring climate suitability with USDA Data. I presented it at Thomas Jefferson Symposium to Advance Research.
Flask, React, JavaScript, SQLite
Social Media Platform Prototype
I developed a prototype social media platform that supports user profiles, authentication, posts, comments, and likes. I built the pages with JavaScript and React and used a Flask server for authentication, routing, and error handling. I utilized Rest APIs and Ajax Calls for communication with the server and an SQLite database for storage.
React, JavaScript
Michigan Computer Graphics Club Website
I developed and maintained the club website with React, JavaScript, and Chakra UI. This website significantly increased the club's online visibility and facilitated communication and engagement for its 200+ members.
Python, OpenCV
OpenCV Security Camera
After encountering uninvited guests in my house in 2022, I developed a Raspberry Pi Security Camera to detect and record motion and detect, record, and classify audio. Its functionality is provided by OpenCV’s MOG2 and KNN models and YAMNet and PyAudio models respectively. I also surveyed potential users on security cameras preferences to better suit their needs and use cases, with support provided by the Michigan Alumni Network and Michigan Center for Entrepreneurship.