On the JEPA team for video understanding.

Research on understanding the internal function of multimodal models.Advised by Dr. Blake Richards @ LiNCLab.

Researching vision sparse autoencoders (SAEs) and interpretability with Lee Sharkey in MATS cohort.

Researching vision mechanistic interpretability with Neel Nanda. Built the Prisma library, an open source library for vision mechanistic interpretability and sparse autoencoders.

Co-founder and CTO. Transitioned to strategic advisory role after successfully guiding company through initial growth stages.

• Designed and implemented an end-to-end analysis pipeline for receptive field analysis from calcium imaging data in mouse visual cortex at the Stringer Lab• Created internal Python library and GUI for receptive field analysis and exploration• Built a versatile tool that researchers can use through a visual interface or code• Conducted comparative analysis between mouse visual cortex data and optic flow neural nets.• Specialized in deep learning and neuroscience data analysis, scientific software development, and computational modeling of visual systems

Cohort 6

• Led intern research team in evaluating and benchmarking GPT models trained on specialized datasets• Optimized pipeline accuracy by 10% through dataset curation and benchmarking• Specialized in LLM training, pretraining, and evaluation techniques• Conducted novel research on improving LLM training methods, including exploration of RL-based approaches and tokenizer optimization• Engineered critical components of a RAG-like system for long-form document retrieval, serving banks and B2B clients• Optimized database latency, improving stack efficiency by 30%• Contributed to core organizational processes and published research findings on company research blog

- Wrote thesis on semantic representation in the human brain in collaboration with the Princeton Neuroscience Institute and Google. - Leveraged knowledge of data science and machine learning, including methods like PCA and multivariate regression, and Python libraries like Tensorflow, NumPy, scikit-learn, and pandas, to analyze neural response.

Leveraged knowledge of dynamical systems to build models of the homebuilding market.

Analyzed financial health of companies for Princeton's only investment organization that donates returns to charity.

- Built database of ethnographic information for the Natural History of Song Project.- Implemented web scraper saving 15 hours/week.

- Made executive decisions for student-run three-day tech conference on nanotechnology, artificial intelligence, genetic engineering, VR/AR, and nuclear fusion, attended by 154 funded student entrepreneurs, award-winning STEM grads and undergrads, and published PhDs from top universities.- Led campaign that increased membership from 70 to 250+ within one year, and organized the recruitment and orientation of 20+ directors.- Ran strategy to promote artificial intelligence safety to international conference attendees and the Princeton student body.

Designed a pilot study on the correlation between air quality and mood in Boston, analyzed data statistically, created air quality activities and presentations for museum guests, and presented results of study in an interactive public forum of policymakers, city planners, scientists, museum staff, and citizens.