Xiaomeng Li

Co-founder & CTO, NanoRhino · Houston, TX

About

I build and study production AI coaching systems. I'm the co-founder and CTO of NanoRhino, an AI nutrition coach that works entirely over text messages and is paid only on results — $10 per pound actually lost, capped at $500.

My current work treats weight coaching as a control problem with a dirty sensor: self-reported food logging carries a large, person-specific bias, so the system maintains two energy-expenditure estimates in deliberately different units and routes between fixing the target and fixing the measurement. The controller is deterministic; the language model composes conversation but never computes a number a user sees. This architecture runs in production across hundreds of long-lived, per-user agents.

Before founding NanoRhino, I was a software engineer at Amazon (Vancouver) and Baidu (Beijing).

Research

Two parts of the NanoRhino coaching engine are U.S. patent-pending: the deficit-loop calorie controller described above (provisional application filed September 2026), and the per-user dish memory for food-energy estimation (provisional application filed October 2026).

Selected writing

• Oct 2026We Tested Our AI Calorie Estimator on Google's Nutrition5k — 455 weighed dishes; portion size is most of the error.
• Oct 202669% of Our LLM Bill Was Cache Writes — prompt-cost engineering for a per-user agent fleet.
• Sep 2026Getting Paid Only When Users Lose Weight — what results-based pricing does to engineering.
• Sep 2026Why Your TDEE Is Probably 500 Calories Too High — what 29 measured users showed us.
• Sep 2026The 10 Weight-Loss Myths Real People Actually Believe — from 7,837 coaching messages.
• Aug 2026The Deficit Loop: Why Your Calorie Target Should Change Every Week
• Jul 2026How Accurate Is AI Calorie Tracking? How We Cut Our Error to 3.8%

Experience

Education