At Zapier, I built the AI Zap Builder evaluation harness that became the gate for model and prompt changes, raising offline-to-online metric correlation from ~0.4 to 0.82. I also architected RAG infrastructure powering Copilot, AI Zap Builder, and Tables field enrichment.
I designed a LangGraph multi-agent orchestration system that backs four production AI features, and authored MCP servers exposing Zapier's app integrations as agent-callable tools. I also instrumented LLM observability and migrated ML training and serving templates to a unified platform.
On Zapier's Editor AI team, I shipped the retrieval and ranking system behind Copilot from beta to GA, serving 100K+ daily active users. I established evaluation frameworks and owned the ML lifecycle for the trigger/action recommender, which drove a 19% lift in zap completion rate for new users.
At Shopify, I led ML modeling for merchant matching and developed demand forecasting for fulfillment capacity planning. I also productionized a contextual-bandit personalization engine and built NLP pipelines for merchant product descriptions.
