Sullivan Lee

Labeled sound events in videos for a Meta AI audio project, marking precise start/end times and short descriptive labels while maintaining high accuracy and consistently clean annotations.

Annotated audio, video, text and geo data for ML training across 1,000+ datasets, maintaining a 99.9% QA accuracy and building templates that reduced task time by 30%. Created frame-by-frame video object masks and precise sound-event timestamps to improve model performance.

Transcribed aviation audio with speaker labeling and time-stamping, applying domain terminology to meet QA standards and achieve productivity benchmarks while using DataForce and Clockify for task management.

Reviewed and flagged issues in AI-generated responses for chatbot and LLM accuracy, assessing over 100,000 prompts and responses and rewriting internal annotation guides to cut onboarding time by 40%.

Created 500+ prompt–response pairs from website screenshots to train language models on UI understanding, maintaining a 98%+ approval rate and delivering work on or ahead of schedule.

Transcribed legal recordings including depositions and hearings with strict verbatim standards, delivering formatted transcripts and weekly reports that surfaced LLM response issues and unusual behavior patterns.

Pursuing a Bachelor of Science in Data Science with coursework in machine learning, data visualization, programming for data science, statistical methods, natural language processing, AI ethics, and research methods.