San Francisco Bay Area
Machine Learning Engineer and Applied Science leader and hands-on builder who thrives at the intersection of applied research and business impact. With a PhD in applied mathematics (statistics & ML focus), multiple patents, and peer-reviewed publications, I bring a researcher’s rigor to the messy realities of production systems. In my 10 years in industry, I’ve turned ambiguous, high-stakes problems into scalable solutions that combine statistical innovation, AI/machine learning, and pragmatic engineering. Through these projects, I've delivered results across ads/marketing, energy, finance, defense, and AI.I thrive in cross-functional, heavily collaborative environments. I’ve led cross-functional teams, shaped product strategy, and navigated complex, multi-stakeholder landscapes while continuing to develop novel methodologies ranging from attention-based models for large-scale audience measurement to privacy-safe causal inference pipelines to end-to-end forecasting and trading platforms/strategies.
Machine learning and data science consulting with successful projects in: * Energy (ML) * Finance (Statistics) * Media (ML+Statistics) * Real estate (ML) * AI (Graph RAG; AI evaluation; Representation learning) * Marketing (ML+Statistics) * Games (Reinforcement Learning) * Toys (Model distillation and RAG)
Designed, built, and deployed an end-to-end data pipeline, analysis platform, feature store, backtesting framework, trading strategy, and human-in-the-loop trade recommendations for PJM westhub day ahead and real time instruments (power futures) trading on ICE. Modeled the physical systems of PJM, and refined and implemented a number of network analysis algorithms in support of FTR (financial transmission rights) trading.
• Owned the Exposure Measurement domain across 4+ products and led cross-functional DS/Eng/PM teams from prototype to production, influencing product, C-suite, and top-tier client decisions. • Personification tech lead: designed, built, and shipped household→person-level modeling that unlocked tens of millions in ARR; served as mission-critical on-call. • Out-of-Home measurement (0→1): launched a new product by fusing novel signals with an Empirical-Bayes estimator; materially expanded top-partner engagement. • Designed and prototyped a unified embeddings/representation layer leveraging rich viewership signals to improve downstream ML beyond noisy metadata. • Scale & performance: integrated/optimized Snowflake UDFs to run measurement on 10B+ rows; developed a PyTorch set-transformer that improved hold-out accuracy for person-level metrics. • Methodology: developed a correlation/reach model that balanced stakeholder needs; introduced LLMs into DS workflows; adapted methods to privacy-safe cleanrooms without sacrificing statistical validity. • Leadership & mentorship: grew and coached DS/Eng talent; interviewed senior candidates; partnered directly with executive-level clients and data vendors; built financial models that informed data-provider pricing/selection. • Patents (pending): Modeling & Personification in a Cleanroom; Outcome Measurement in a Cleanroom; Out-of-Home Media Measurement. Tech: Python, PyTorch, LightGBM, Spark/Databricks, Snowflake (UDF/Snowpark), SQL; privacy-preserving cleanrooms.
• Designed and trained a multi-task TensorFlow model with a shared trunk and auxiliary prediction heads over TV panel data. The auxiliary tasks (other TV viewership outcomes) were used purely as auxiliary losses to improve the shared representation and regularize the model, significantly boosting performance on the primary ad-exposure prediction task. Used in production to improve cross screen reach for a top 10 client. Described by client as a "massive success." • Primary analyst and quantitative developer for online-to-offline ad impact causal measurement product – Develop statistical methodology and a large pipeline and report generating framework with illustrative, interactive visualizations to assess propensity model quality – Lead and deliver analyses that direct major product decisions and launches – Collaborate with product, engineering, and third party data vendors to develop consensus on statistical methodology and messaging for cross-media sales lift measurement sub-product • Co-lead Google’s Just Cause Python causal inference package • Contributed to multiple open-source software projects including TensorFlow and CoLaboratory • Advised multiple Google X projects on causal inference techniques, assumptions, and interpretations