Wooyoung Son

I had an exciting opportunity to work on a project that involved re-engineering a mobile application prototype. The goal of the application was to detect and screen vision problems in children and refer them to optometrists for further testing and treatment. I proactively analyzed the current system and gathered functional requirements, using various UML diagrams to redesign the architecture and refactor the codebase. I employed the Inversion of Control and Single Responsibility principles to decouple a lot of the code, making the application maintainable and production-ready.

Working with the back-end team, I implemented a data uploading feature to an AWS server for real-time data collection. This data could be used by researchers to improve vision screening practices. I also added a distance estimation feature using a machine learning algorithm to provide appropriate feedback to the user, advising them to maintain a proper distance from the mobile phone to optimize vision screening.

To ensure security and data protection compliance, I configured AWS Cognito User Pool for user authentication and wrote a Node.js module to verify access tokens and restrict access to protected resources.

I created a plugin that enables script writing without server restarts, which increases server reliability by supporting hot replacement and allows for the implementation of simple functionalities without the need for writing Java code.

As an administrator, I know firsthand that writing Java code can be challenging, and writing simple scripts in the custom programming language I created can be much more productive. Additionally, this plugin's language is not dependent on strong typing like Java, making it easier to integrate multiple plugins. With this plugin, I was able to reach production much faster and provide a more efficient workflow.

During a school project, my team and I developed an inventory tracking system. The system allowed users to read information from plain text files, including media stock information, user information, and miscellaneous information.

My role in the project was to create a hashtable that allowed for constant time access to each inventory item, and I also conducted unit tests to ensure the quality of the software. The project was a great opportunity to work collaboratively and learn about the importance of testing and efficient data structures. Overall, I enjoyed working on the project and was proud of the final result.

As part of a school team project, we developed a 2.5D exploration game with horror elements where the player character needs to find a way to escape while a monster is chasing them.

As one of the team members, my main responsibility was to organize the code produced by my teammates and encourage them to produce modular code for easier integration in later stages of development. I spent a lot of time refactoring the code while ensuring that all the functionalities were maintained.

Additionally, I implemented a general-purpose Finite State Machine that drove the monster's behavior in real-time, making the gameplay more exciting. I also developed the user interface elements for the game, which showed the current progress of the user's objective. Overall, my contributions helped to improve the game's functionality and user experience.

I developed a prototype software that aims to help marketers quickly write blog articles that are likely to receive more exposure. The process involved web scraping using Python's BeautifulSoup4 and requests library, with data stored in SQLite.

The data was then queried, pre-processed using Open-refine software, and fed into a pre-trained GPT-2 model that was fine-tuned for five epochs on the entire dataset. The resulting output was evaluated and deemed convincing by a human evaluator.

I collaborated with a professor and her research team to develop a machine-learning model that could detect brain hemorrhages from ultrasound images. We analyzed the ultrasound image by dividing it into 12 sectors and combining them using various methods like sum, average, and median. We then analyzed them using Principal Component Analysis.

As an additional experiment, I was assigned to retrain the model developed by a previous graduate student to see if it would perform better at different time frames of the ultrasound image.

Our aim was to improve the model's accuracy in detecting hemorrhages by identifying the specific time frame within which the ultrasound image showed the best sign of hemorrhage. We tested the model's performance on different time frames. We found that the accuracy was better in the specific frame used than in its neighboring frames, indicating that the experiment was successful in improving the model's accuracy.

I developed a web-based platform to assist marketers in finding popular items for sale. The platform uses metrics such as Google Trends and the number of items available on online shopping sites to provide users with valuable insights.

I built the platform using React/Next.js for the front-end, Django/DRF for the back-end, and Maria DB for the database. Additionally, I created several Python scripts to scrape websites and process data in the database.

To ensure easy deployment to cloud environments, I designed the entire system using Docker. With this set up, if the software proves profitable, it can be easily transferred to cloud environments.

I was tasked with implementing a deep learning algorithm for text classification for a startup's machine learning platform. I chose to use GPT2 since it was a state-of-the-art algorithm with fewer parameters, making it suitable for the client's needs.

Using a smaller model also helped speed up the development process, and I ensured the code was cleanly written to work with the provided specifications. The client was pleased with my expertise and knowledge in AI/ML tools, as well as my clear communication and timely delivery. Overall, I received high praise for my work on this project.

"Wooyoung is a great AI/ML engineer. He delivered a GPT2-based text classification to us as per our specifications. He is very knowledgeable [and] skilled in AI/ML tools. He communicates clearly and delivers on time. Highly recommend working with him!"

I developed a Node.js software application that leverages the Puppeteer library to automate feed searching and execute various configured tasks. The application uses a state machine agent that continuously transitions between states depending on the browser context and adapts to unexpected variables.

I opted for a state machine design due to the chaotic nature of internet communication and unstable network connections that make it necessary to discretize variables into finite states to better handle such variability.

This approach also makes it easier to maintain the existing codebase and add new features. Overall, I am proud of the state machine-based design of this project, as it has allowed me to create a robust and adaptable software application.

I developed an inventory system for games using Unreal Engine. I utilized C++ to implement the inventory logic, including adding, removing, and querying items, as well as locating empty spaces. Additionally, I made the inventory abstract to accommodate a specialized form of inventory UI, ensuring that user interaction with the inventory would be less impacted by UI changes. The blueprint script was utilized to perform the remaining steps, which natively accessed the C++ code through Unreal Engine's serialization support. With this approach, I efficiently developed a stable inventory system for the game that is unlikely to undergo core logic changes.