Edwin He

GPA: 3.71

Coursework: Machine Learning, Artificial Intelligence, Computer Vision, Data Structures and Algorithms, Databases, Operating Systems, Networking, Computer Architecture, Python, C++, Java

• Fine-tuned Large Language Models (LLMs) using RAG (Retrieval-Augmented Generation) to ensure medical advice remained grounded in verified clinical data
• Optimized LLM inference for usage in low-bandwidth environments common for underdeveloped regions
• Orchestrated a multi-agent AI workforce to automate code review, documentation, and localization, streamlining the adaptation of health advice for different dialects in target countries

• Supported students with C++ programming concepts including data structures, memory management, and algorithmic efficiency
• Guided students through debugging processes, improving understanding of pointers, dynamic allocation, and STL
• Collaborated with instructional staff to review coursework, clarify conceptual challenges, and ensure consistent grading standards

• Annotated and reviewed large datasets for training and evaluating LLM programming capability, ensuring data accuracy and relevance
• Tested model outputs for correctness, consistency, and alignment with task instructions
• Provided detailed feedback on model performance, including edge cases and anomalies

• Built a PyTorch-based training pipeline using 3D motion capture data and audio data, modeling root translation and joint rotation for audio-based full-body dance synthesis
• Converted 3D model hierarchy and motion channels into normalized tensor representations and aligned them with audio-derived features from audio files
• Trained temporal sequence models on synchronized audio–motion windows to generate beat-aligned dance motion

• Reduced stutter frametime from 42ms to 7ms by optimizing file loading and processing
• Achieved microsecond note and audio synchronization using absolute time calculation system with sub-millisecond precision
• Architected cross-platform file management system supporting Windows, Linux, macOS and web with persistent per-song settings and statistics