DoorDash - Machine Learning Engineer - August 2024 - Present

• Developed a multi-modal RAG model (text & visual features) using SOTA models to create item-level tags, driving a significant increase in sales.
• Created an auto-prompting methodology to improve LLM accuracy for structured data tasks, reducing ML time by approx. 80% while increasing precision by 16%. Presented the results to the broader ML team and leadership.
• Deployed LLMs as batch jobs integrated with a CRDB database via Kafka pipelines, and constructed the supporting data pipelines with logging, monitoring, and alerting.
• Developed & deployed causal ML models (S-learners & double machine learning) to generate recommended actions for merchants.
• Trained a multi-head neural network using Fireworks and evaluated it against existing LightGBM S-learners.
• Conducted technical interviews, assisted with road map planning, and partnered with cross-functional teams.

Gopuff - Data Scientist II - July 2022 - July 2024

• Developed an end-to-end in-house switchback platform with a reusable metric infrastructure, performing all tasks from registration to analysis. Built the front-end using Streamlit, deployed with Docker on Azure using FastAPI, with results provided as an MLflow dashboard and unit tests in pytest.
• Created a full test suite & provided education for the switchback platform, allowing non-DS users to create switchbacks. Over 80 switchbacks were registered on the platform, acting as the experimentation component in reducing delivery costs by 40% over 18 months.
• Averted several long-term negative feature roll-outs by developing the first long-term value (LTV) model of customers using double machine learning and surrogates. Productionalized the model, provided LTV results automatically alongside A/B tests, and presented the project to leadership.
• Assisted with experiment design, power analysis, & estimation of price elasticities, leading to a 4% margin lift with little order impact — the largest company-wide margin positive initiative.
• Developed supply chain experimentation tools and onboarded supply chain users to the experimentation platform, determining the appropriate methodologies, metrics, and time horizons for such experiments.
• Reduced expiry costs by 3.5% and decreased manual workload by improving the LightGBM based product forecasts using product promotion features.
• Used diff-in-diff & synthetic control analysis to evaluate the effectiveness of marketing initiatives.

UCLA Anderson - Research Assistant to Andrea Eisfeldt - 2018

• Assisted with empirical finance projects, responsible for data sourcing, data cleaning, and empirical model development.

Job Market Paper

• The Heterogeneous Effect of Local Land-Use Restrictions Across US Households. _{Abstract: Using a structural approach, I quantify the effect of land-use regulations on different age and education groups. Building on the seminal work of Roback (1982) I estimate a dynamic spatial structural equilibrium model of household location choice, local housing supply and amenity supply. I show that in the long-run, removing land-use restrictions benefits all household groups and increases aggregate consumption by 7.1%. These consumption gains vary across households, less educated and younger households see increases in consumption about twice as large as more educated or older households. In contrast, in the short run, removing land-use regulations reduces the consumption of older-richer homeowners while increasing the consumption of younger renters. In a counterfactual 1990-2019 transition abolishing land-use regulations reduces the consumption of households born before the mid-1960s, while increasing consumption of more recent generations. Given the difficulty in reforming land-use regulations, I explore whether a shift to remote working or creating new urban areas leads to similar consumption gains compared with removing land-use restrictions. Qualitatively I find the gains are similar, but quantitatively are only about 20% as large as abolishing land-use regulations from existing urban areas.}

Publications

• Corporate Bond Liquidity during the COVID-19 Crisis.
  (with Mahyar Kargar, Benjamin Lester, Shuo Liu, Pierre-Olivier Weill, and Diego Zúñiga)
  Review of Financial Studies, 2021
  For a non-technical summary click here

Work in Progress

• Trading Relationships in the US Corporate Bond Market
  (with Diego Zúñiga)

Languages
Highly proficient in Python, SQL, R, and LaTeX. Working knowledge of Julia.

Developer tools
Docker, Fireworks, Uvicorn, FastAPI, MLflow, PySpark, Git, Streamlit, Azure, Databricks.

Libraries
pandas, poetry, PySpark, numpy, EconML, XGBoost, LightGBM, Statsmodels, scikit-learn, PyTorch.

Instructor

• Econ 106F: Finance (UCLA Summer 2019 and Summer 2020).

Teaching Assistant

• Econ 1: Introduction to Microeconomics (UCLA - Winter 2018)
• Econ 2: Introduction to Macroeconomics (UCLA - Winter 2020)
• Econ 101: Intermediate Microeconomics (UCLA - Fall 2018)
• Econ 102: Intermediate Macroeconomics (UCLA - Fall 2017, Spring 2018, Winter - Fall 2019, Winter 2021, Spring 2021)
• Econ 103: Econometrics (UCLA - Summer 2021)
• Econ 106F: Finance (UCLA - Fall 2020, Spring 2020)