Instacart: growth, ads, and marketplace ML
Production ML across growth targeting, ads bidding, inventory availability, fulfillment decisioning, and marketplace systems.
Resume detailsPhD Economics, UCLA · Senior machine learning engineer
Economics, machine learning, and marketplace algorithms.
I build and study products where measurement matters: experiments, causal inference, marketplace algorithms, production ML, and decisions under uncertainty.
Projects
Selected work.
Short version: economics training, production ML, real marketplace systems, and independent AI products.
Microsoft Research: search auctions
Worked with Susan Athey on Bing search-auction counterfactuals, ranking rules, and marketplace revenue effects.
Research backgroundOpenClawBrain
Local agent memory and evidence-based continuity: what an AI assistant should remember, when to trust it, and how to prove what changed.
Product siteBountiful Garden
Neighborhood produce-sharing marketplace. Product, trust, local coordination, and full-stack execution.
Live productProject Pelican
Private autonomous research workflow for options: data, forecasting, risk controls, execution, and monitoring.
OverviewIssued patent
Automated policy-function adjustment using reinforcement learning, tied to production bidding systems.
Patent PDFInstacart
Growth, ads, and inventory systems.
Three areas where the work combined economics, production ML, and real marketplace feedback.
Ads Automated Bidding
Built Instacart's first optimized ads automated bidding system from scratch. Replaced manual product-keyword bids with weekly-budget bidding across roughly 70% of Sponsored Products spend, about $700M annualized.
Technical writeupGrowth Systems
Built the MVP for growth targeting from causal experiment outcomes. Ranked users by expected incremental treatment effect instead of naive conversion propensity.
Technical noteInventory Intelligence
Built item-store availability and fulfillment ML across noisy catalog, store, shopper, and operational signals. Work included real-time LightGBM serving, calibration, rollout, monitoring, and critical-item prioritization.
Resume detailsPDFs
Resume and papers.
The public PDF surface is intentionally small.
Writing
Technical explainers.
Long-form notes on technical architectures and marketplace ML.
Building an Automated Bidding System for Marketplace Ads
A public-safe technical writeup on moving from manual product-keyword bids to objective-driven automated bidding, with return estimation, budget pacing, bid simulation, and measurement.
Read the writeupGrowth ML: causal incentive allocation
A public-safe technical note on randomized identification, uplift modeling, offline policy evaluation, constrained optimization, and budget pacing for growth ML.
Read the noteTemporal Fusion Transformer: architecture, intuition, and Pelican walkthrough
A visual guide to TFTs with the actual Pelican usage path: feature families, labels, candidate scoring, and a real AAPL spread decision.
Read the explainer