Focused on applying advances in mechanistic interpretability to the problem of machine unlearning, which addresses recent concerns over LLMs adherence to copyright law by enabling the removal of PII / copyrighted content from the model without compromising overall utility.

• Achieved a 19% reduction in Word Error Rate (WER) for ArtFair's Caption-Generation service through speaker-specific finetuning and diarization enhancements• Diagnosed key failure modes of the base ASR model on YouTube client audio data, including overlapping speech, domain-specific terminology, inconsistent punctuation, proper nouns/gamer tags, and background noise, quantifying their impact on WER.• Curated and standardized datasets to fine-tune OpenAI's Whisper-Large-V2 ASR model and Pyannote's Segmentation-3.1, specifically addressing the diagnosed failure modes and improving target-speaker performance.• Delivered a live demonstration of system improvements to YouTube client channel stakeholders, showcasing an average 4-hour reduction in manual caption correction time and a 66% reduction in captioning overhead.

• Developed a question generation system for CulturalCSBytes (CCSB), a tool designed to make computer science education more inclusive and relatable.• Applied advanced language models to autonomously generate coding questions, merging core CS principles with diverse academic subjects like World History, Music, and Gaming.• Designed a user-friendly command-line interface, enabling educators to generate questions tailored to a variety of learning objectives; Contributed a comprehensive Jupyter notebook, offering insights into the project's inner workings and facilitating resource reproduction or adaptation.• Generated a final set of over 800 questions across 12 domains (math, music, athletics, history, business, everyday life, school/career, art/design) to be used in CCSB system.• Collaborated with educators and stakeholders, ensuring the project's alignment with educational goals and achieving a significant impact in making CS concepts accessible across interest areas and academic disciplines.

• Developed system using Python, JavaScript, Selenium Webdriver, and Instagram/Twitter APIs to capture trends in Name, Image, and Likeness data from across NCAA College Athletics.• Compiled accurate set of athlete social media profiles in order to determine social media marketing value and to power future monetization efforts; Developed novel disambiguation algorithm.• Operationalized for scalability and maintainability.

• Directed the expansion of OpenBrackets' educational programs from the Bay Area to Seattle.• Recruited and managed a team of qualified high school students from across the district to teach OpenBrackets courses.• Led student outreach campaigns in the Seattle Public Schools district, presenting to and forming partnerships with the Equity, Partnerships and Engagement administrative division as well as multiple SPS schools and teachers.• Taught and TA'd for Python programming courses in fall 2020 and spring 2021.• Assumed role as acting Vice-President for summer 2021 to direct OpenBrackets' Summer Workshop Series and develop plans for Fall 2021 programs.

• Achieved granular control over data preparation and analysis in Microsoft PowerBI and multiple Python libraries.• Generated data visualizations for Affiliate enterprise marketing program to determine parameterized sales patterns.• Incorporated work into reporting of Affiliate Program for performance monitoring and anomaly detection.