Yueheng(Bryan) Wu

Data Science & Computer Science Student

A bold, solutions-driven Data Science undergraduate passionate about using machine learning and system design to solve real-world problems.

About Me

I'm Bryan Wu, a Data Science freshman student about applying machine learning and system design to solve real-world problems. I thrive in fast-paced, high-stakes environments where impact matters and enjoy driving projects from concept to execution.

Built SPYN as the Sole Backend Developer with three teammates in 36 hours at PennApps XXVI—an AI-powered posture and exercise coaching system for laptops and phones, providing real-time feedback to prevent diseases and improve performance.

Placed in the Top 3% (43/1,449) in the Kaggle Eedi "Mining Misconceptions in Math" competition, where built and fine-tuned LLM pipelines with GPT-4 data generation and contrastive learning.

Authored a research paper on real estate price prediction, accepted at ICFMDE 2024.

Actively participate in hackathons, volunteer programs, and leadership roles, bringing together technical expertise and strategic communication.

Skills & Technologies

PythonMachine LearningData VisualizationSQL

Education

Data Science, B.S.

University of Wisconsin–Madison

2025 - 2028

Location

Madison, WI

Open to remote opportunities

Projects

SPYN - AI-Powered Posture Coach | PennApps XXVI

My team built an AI-powered posture and exercise coaching system that runs on laptops or phones, providing real-time feedback to prevent diseases and improve performance.

My Contributions (Sole Backend Developer):

•Research & Scoring Framework: Conducted comprehensive biomechanical research to establish scientifically-grounded scoring standards with precisely calibrated weight distributions for posture and exercise analysis.
•Computer Vision System: Built multi-modal scoring pipeline using MediaPipe with custom geometric algorithms, 6-dimensional posture analysis, and intelligent penalty mechanisms.
•AI Architecture: Designed and implemented dual-AI integration (Cerebras LLM + Gemini AI) with confidence-based dynamic weighting for real-time pose optimization and session-level intelligence.
•Session Intelligence: Created session aggregation system with time-weighted scoring, deduplication mechanisms, and intelligent throttling controls for personalized recommendations.
•Backend Infrastructure: Implemented real-time metrics pipeline (JSON persistence + REST APIs) with session aggregation, time-weighted scoring, and intelligent throttling for scalable deployment.