Mian Luo

I’m a senior software engineer at Pinterest, building the AI behind visual shopping, search, and ads—turning loose requests into real outcomes with an agentic shopping assistant and LLM-powered retrieval and relevance.

My work has supported high-scale usage (the assistant handles 400k+ requests/day) while improving key metrics, including roughly an 11% lift in shopping-session conversion and about a 9% lift in ad click-through via LLM-generated and scored ad-copy variants.

Before Pinterest, I spent 8 years at Google on search and ads serving, backend infrastructure, and release engineering—owning systems end-to-end and focusing on performance and reliability. I automated the build, canary, and staged-rollout pipeline, cutting deployments from most of a day to under an hour, and reduced p99 latency on a core serving RPC by about a third.

I’m comfortable owning a system from model and API down through serving, scaling, and on-call. I build with batched inference, embedding caches, fallback and routing under load, and I ship with rigorous evaluation—offline relevance graders, LLM-as-judge checks, and online A/B tests—while mentoring engineers and establishing reusable backend design patterns.