Dec 2023 – Apr 2024 · 4 mos

• Co-lead two teams working on medical AI, foundation models, representation learning, and federated learning. Recent work includes co-proposing and leading the Brain Moonshot behind Med-PaLM, a large language model (LLM) aligned to the medical domain, published in Nature, deployed to healthcare organizations via Google Cloud, and featured in The Economist, Scientific American, Forbes, CNBC.

Apr 2023 – Nov 2023 · 7 mos

Oct 2021 – Mar 2023 · 1 yr 5 mos

Oct 2020 – Sep 2021 · 11 mos

Sep 2019 – Sep 2020 · 1 yr

Jun 2018 – Sep 2018 · 3 mos · Mountain View

• Used Image Search query data and other unsupervised web data to learn multilingual word embeddings
• Improved performance 77% over a text-only baseline by using image data as a bridge between languages
• Built infrastructure to rapidly train and evaluate using internal and external datasets

Jul 2017 – Mar 2020 · 2 yrs 8 mos · San Francisco Bay Area

• Cofounded and instructed Stanford's first AI ethics class "AI for Social Good" (cs21si.stanford.edu) for two years (gave lectures, made course material, invited guest lecturers)
• Organizer for CS50, in which teams of coders work on high-impact projects with nonprofits
• Cofounded and taught new AI+Social Good workshop series – first workshop had 50+ attendees, 400+ online viewers
• Teaching assistant for ENGR245: The Lean Launchpad and CS224N: Natural Language Processing with Deep Learning

Jun 2017 – Aug 2017 · 2 mos · New York City Metropolitan Area

• Built several critical components of a platform that runs deep learning algorithms on a livestream
• Developed NFL team image classifier with 98% accuracy, reducing prediction time from 7.8s to 0.028s
• Created a module to parallelize model prediction 10x quicker than Python built-in
• Devised a protocol to speed up image transfers 3-5x compared to NumPy’s built-in
• Built a web application that simplifies training models for the platform

Nov 2015 – Jun 2016 · 7 mos

• Worked with Stanford classmates on an early-stage market research startup
• BASES Challenge Finalist (Stanford’s premier startup competition)
• Conducted over 100 interviews and experiments with marketers and consumers to iterate product
• Led development and testing of product prototypes and conversion funnel
• Admitted to Steve Blank’s selective Lean Launchpad program as a freshman (usually reserved for grad students), returned the next year as a TA
• Data Does Good is a Public Benefit Corporation that helps people turn their personal data into cash for their favorite nonprofit.
• The problem: Unlike tech companies who are drowning in data, traditional consumer-facing enterprises (CPGs, Restaurants, etc) are forced to understand their customers through market research and surveys. These surveys are limited to what a consumer can recall and don't come anywhere close to the depth of insight from actual behavioral data (event streams, transactional data, etc).
• We created Data Does Good as a way to get consumers to instrument their devices and apps to share their behavioral data so that we could turn that data into insights for our customers.

Dec 2012 – May 2016 · 3 yrs 5 mos

• Provided website design, development, and marketing services pro bono for several small nonprofits
• Partnered with eye-tracking startup Viomba to increase donation rates by 7x for a NC free clinic