Career

Experience

1. 2025 – present

   Research Scientist · GranicaResearch

   Working with Chief of Science Andrea Montanari on approximate leave-one-out methods for non-convex learners and on Large Tabular Models.

2. 2018 – 2025

   Scientist and Engineer · Google / YouTubeFounding Eng

   • Founding engineer on YouTube Shorts; designed dozens of recommendation algorithms.
   • Replaced deterministic Shorts shelf placement with a reward-model-driven probabilistic policy, implemented in production C++.
   • Core contributor on the Gemini 1.5 Technical Report: self-critique LLM evaluation and model-assisted estimation combining human and LLM scores.
   • Implemented a Content-Aware Degradation-Driven Transformer for video super-resolution in JAX.
   • Awarded readability in C++, Python, R, and SQL; readability mentor in each across Google.
   • Led Technical Infrastructure for YouTube Data Science as the org grew from 50 to 100 scientists.
3. 2017 – 2023

   Lecturer · Stanford ICMETeaching

   Taught graduate Python and C++ (CME 211 / CME 212) to ~200 students per year alongside my industry role. Awarded Best Lecturer in three separate years by students and faculty.

4. 2017 – 2018

   Data Scientist · Cardinal AnalytxIndustry

   First data scientist on the team. Expanded the per-patient feature space from 800 to over 80,000 variables, improving model performance by ~50%. Cut memory and compute by orders of magnitude through sparse-matrix design. Generalized the modeling stack beyond Cost Bloom prediction to arbitrary risk targets (e.g. orthopedic surgery risk).

5. 2017 – present

   Statistical Consultant · Pando PoolingAdvisor

   Authoring core statistical analyses for Pando Pooling, a startup that pools professional athletes with similar projected earnings to reduce individual career risk. Built models over 50,000+ MLB players estimating remaining lifetime earnings, the probability of reaching the Majors, and arbitration likelihood.

6. 2016

   Computational Engineering Intern · Lawrence Livermore National LaboratoryIntern

   Implemented distributed MCMC in Scala/Spark following recent literature (e.g. Neiswanger et al.). Also collaborated with Kaiser Research on ML-based sepsis prognosis.

7. 2012 – 2014

   Research Analyst · The Brattle GroupIndustry

   Worked closely with Armando Levy estimating damages from lost human use of natural resources, supporting the expert testimony of Daniel McFadden on the 2010 Gulf Oil Spill. Authored a half-dozen internal R presentations (e.g. on regular expressions) to lift analyst productivity.