Lead data scientist on Shipt's Personalization team, responsible for the recommender systems behind every personalized shelf on the platform. Mentored 4 data scientists including 2 direct reports.

• Delivered up to 23% Personalized GMV lift across multiple A/B-tested shelves (Trending Items, Similar Items, Deals For You, Complementary Items) over 90 days.
• Designed and built a two-tower deep retrieval model to replace the legacy ALS recommender. The model improved Deals For You with 3–15% lift on Deals-specific shelves and 11% higher click-through rate overall. Offline evaluation showed +406% NDCG, +87% novelty, and +19% long-tail coverage versus the baseline—confirming the model recommends genuinely new, relevant products rather than items customers already know.
• Designed rich user and product embeddings that go beyond surface-level text matching—capturing product descriptions, categories, pricing, retailer info, dietary preferences, and behavioral signals. Built FAISS approximate nearest neighbor infrastructure on GCS to make retrieval fast at scale.
• Created a Personalization Interaction Score that captures the full customer funnel—views, clicks, add-to-carts, and purchases—weighted by funnel depth. This replaced single-metric optimization (e.g., GMV alone) with a composite measure that reflects how customers actually engage with shelves. The methodology has since been adopted by teammates to power next-generation real-time recommenders.
• Developed a price-weighted ATC approach for coldstart users (new customers with no purchase history). Result: 47% engagement lift and 8% more first-time orders over 90 days. The technique was adopted across other shelves after outperforming alternatives.
• Built an automated Shelf Attribution pipeline using fuzzy-string matching to measure Personalization's true contribution to GMV. In the process, identified that previously reported 5% attribution figures were irreproducible—actual numbers sat between 1–4%. Reported the discrepancy transparently, which helped catalyze a shift toward more defensible success metrics across the organization.
• Designed and built a Customer Intelligence Platform (CIP) to centralize customer data, and an "Essentials" recommender shelf for commonly purchased products that serves as both a standalone shelf and a relevance filter for others.
• Partnered with Engineering to replace brittle CSV-based recommendation delivery with Kafka-based pipelines across all legacy recommender systems—a scalable infrastructure improvement that benefits every model on the platform.
• Designed a Retrieval-Augmented Generation (RAG) system to recommend products from Shipt's internal retailer catalogs, and built the business case that convinced stakeholders to invest in an agentic AI framework.
• During the team's first four months, served as the sole data scientist—maintaining all 16 Discovery Science repositories, resolving issues with Search teammates, and keeping the lights on while simultaneously designing next-generation infrastructure.

Core ML engineer at a genomics company developing a blood test for early-stage cancer detection. Worked at the intersection of infrastructure and research, building the systems that scientists depend on daily.

• Key contributor to Freenome's core product: a multiomics cancer detection model that predicts cancer stage (1–4) from blood-draw data. Built data abstractions to handle petabyte-scale genomic datasets, unblocking cross-analyte feature development and meaningfully accelerating training and evaluation cycles.
• Built a model comparison system that tracks research versus production model performance side-by-side—a requirement for FDA audit compliance and a tool that gave the team confidence that production models stayed aligned with research intent.
• Designed and built large portions of Freenome's distributed ML training and serving platform, used daily by 30+ scientists and researchers. Key contributions included scaling CPU-bound training across O(100) machines, supporting multiple evaluation strategies (leave-one-out, K-fold), and building model artifact storage that made reproducibility straightforward.
• Led the adoption of PyTorch, MLFlow, and RayTune across the ML team, replacing legacy tooling to enable GPU acceleration, experiment tracking, and hyperparameter tuning at scale.
• Built a cloud cost monitoring system that surfaced the biggest storage and compute expenses in GCP. The visibility alone drove optimizations that saved the company over $10M annually—making it one of the highest-ROI projects I've worked on.
• Recognized with a Servant Leadership Award, elected by managers and peers across the engineering organization.

Worked on ML systems for health insurance claims processing—a domain where model accuracy translates directly into operational cost savings.

• Designed a ranking model that matches human workers to claims processing tasks based on skill, history, and task complexity. The model delivered $7M in annual value by reducing the volume of manual task assignment and the need for additional hires.
• Built a classification model for partitioning sensitive patient documents, using image and text data to route health insurance claims to the correct processing workflow.
• Developed internal AWS cloud tooling and production API infrastructure to support the ML team's deployment pipeline.
• Led a cross-functional tiger team to prototype a conversational chatbot using Rasa and HuggingFace's NLP library for internal claims inquiry workflows.

Built ML models for an executive recruiting firm, working across the full pipeline from data collection to model serving.

• Developed a suite of models: a classifier to estimate job-departure likelihood, a regression model to predict team sizes from resume features, and a ranking model to surface and match top candidates to open roles using a custom NDCG listwise loss function.
• Built an end-to-end framework for rapid model prototyping, training, evaluation, and serving—enabling the team to iterate on new models without re-engineering infrastructure each time.
• Wrote data collection scrapers to harvest structured candidate data from public sites and APIs.

Two concurrent research positions exploring ML applications in energy and neuroscience.

• California Institute for Energy and Environment (CIEE): Built a recurrent neural network for predicting building energy usage, exploring how temporal patterns in consumption data can inform smarter grid management.
• Bengson Research Lab, Sonoma State: Applied ML models to EEG data to computationally predict individualized occipital lobe activation patterns. The work showed early feasibility for brain-computer interface applications.

Where the foundation was built.

• Data Science Intern, Castlight Health (2017) — Designed an entity matching and deduplication pipeline using gradient-boosted classifiers with hard negative mining. Achieved 85–95% precision/recall across hospital, facility, and practitioner entity types.
• Data Science Contractor, Riviera Partners (2016) — Built a team size prediction model from public data and a Python wrapper for survival model time-series analysis. Set up Flask model-serving infrastructure.
• URAP, Berkeley Institute of Data Science (2016) — Helped map UC Berkeley course progression through different majors by computationally organizing class taxonomies and running deduplication.
• Data Science Intern, Doximity (2015) — Built a gradient-boosted classifier to identify malformed web-scraped articles and used reverse geocoding with fuzzy string matching to link doctor names in news articles to facility profiles.

• BS in Computer Science and Data Science (Dual Degree)
• Berkeley Institute of Data Science Undergraduate Research Apprenticeship (2015)