Sung Oh sunggyeol

Hi, I'm Sung 👋

Computer Science Student at Virginia Tech with a 3.91 overall GPA, pursuing a Bachelor's degree with a minor in Human-Computer Interaction.
Passionate about machine learning, large language models, and their applications in education, AI-driven feedback systems, and real-time data analysis.

Technology on the Trail Lab: Fine-tuned Mistral 7B for an AI-powered writing feedback system, enabling personalized insights on student essays. Engineered backend solutions using embeddings, K-Means clustering, and RAG to analyze writing trends and support professors with data-driven feedback.
IDEEAS Lab: Conducted sentiment analysis and topic modeling on social media data to explore public perceptions of generative AI in education. Combined BERT-based models and LLMs, with LLM achieving 30% higher accuracy. Published and presented findings at the IEEE Frontiers in Education (FIE) Conference, highlighting a 57% positive sentiment toward AI in education.

Developed a high-performance speech-to-speech platform on AWS EC2, integrating real-time transcription using OpenAI's Whisper, WebSockets, and advanced Voice Activity Detection (VAD). Achieved a 60% speed improvement while supporting 100K+ concurrent users with low latency.

Fintellection: AI-powered search engine leveraging LLM embeddings and Retrieval-Augmented Generation (RAG) to deliver real-time financial data insights.
Sentiment Analysis: Analyzed public perceptions of generative AI in CS education, combining BERT-based and LLM approaches, with a focus on thematic and sentiment analysis.
Cloud-Based Speech-to-Speech: Built scalable solutions for speech-to-speech applications during my internship at Fasoo, delivering high accuracy and performance.
Diary Study Platform: Developed an innovative diary study platform that leverages fine-tuned large language models to analyze qualitative diary entries. Using fine-tuned Mistral 7B, the platform extracts thematic trends and sentiment nuances from unstructured data.
VT Copilot: AI-powered campus assistant using LLM embeddings and RAG, featuring a data analyst agent that turns coursework-related data into interactive charts. Built with FastAPI and React for seamless user interaction.
Internal Deep Research: Designed an AI-powered research assistant that generates structured reports from internal documents using LLMs and FAISS-based retrieval. The tool recursively scans text and markdown files, aggregates content, and synthesizes iterative markdown reports with citations.