Work

Four systems from my time at Pattern. Each has a short summary and a longer story.

• Up to ~20M predictions a day at peak

  Conversion-rate prediction for ad bidding

  The models behind live ad bids on Amazon, Walmart and Tmall, and the data checks that stop a bad run before it reaches the bids.

• Pre-production to live pilot

  An LLM customer-service agent

  I took over and reworked an agent that answers product questions from text and photos, and tested every change against real customer questions.

• ~600k simulations, ~40 minutes

  A decision engine for excess inventory

  I designed and built it, handed it on, and was brought back to own it. In 2026 I cut its compute credits by over 96%.

• In production since October 2026

  An MCP server for AI agents

  Lets the company's internal AI agents answer inventory questions, built so an agent cannot sound more certain than the data.