A sci-engineer passionate about building consumer and enterprise software products.
Experience
2025 — Now
2025 — Now
San Francisco Bay Area
Grazing LLMs on organic, non-GMO document corpora
2021 — 2024
2021 — 2024
San Francisco Bay Area
Tech lead for the following areas:
• MLflow (mlflow.org), a unified end-to-end open source MLOps platform for GenAI applications with 0.6M daily downloads (Pytorch at 1.2M for comparision)
• MLOps Development team, serving thousands of data scientists developing and deploying ML applications for enterprises of all kinds (from family businesses to Fortune 500)
• Model Serving network and compliance, securly and reliably serving thousands of model endpoints from enterprise customers
2015 — 2021
2015 — 2021
Mountain View, CA
Lead the Loon simulation and system evaluation team, managed a group of high-performing software engineers and data scientists, provided Evaluation-as-a-Service to several critical functional groups at Loon, including business development, system engineering and fleet planning and control.
Designed a generic large-scale robotic simulation and evaluation platform (codename RoboEval), performing end-to-end system simulation, evaluation, validation and verification for fleets of autonomous flight vehicles. RoboEval produces and analyzes PBs of data on a regular basis with high efficiency (X millions compute budget saved annually), provided
• High-fidelity and statistically robust revenue and cost analysis for business deal-making.
• Design space exploration for next-gen vehicles via grid scan and automated parameter search.
• A/B testing framework for evaluating ML models for fleet planning and control systems.
Designed and implemented aerial LTE network simulation and analysis service, with demand / supply optimal matching and steerable beam targeting. The service is central to production network ops and business intelligence, saved licencing cost Y millions per year.
2013 — 2015
2013 — 2015
Improving YouTube homepage user engagement.
Responsibilities include:
Design / implement / evaluate new features / products for YouTube browse pages;
Train and serve machine learned systems at YouTube scale;
Analyze user data for better personalization strategy.
Tech lead on the following YouTube products:
YouTube Music homepage browse infrastructure.
My Mix -- a machine learned radio station, serving fully personalized music to millions everyday.
Designed and deployed onboard Lidar object detectors to Waymo vehicles.
2010 — 2013
2010 — 2013
Theoretical modeling and developing computational methods in quantum chemistry.
Proposed the world's first molecular Wankel engine operating at THz with infrared radiation. Highlighted in Nature Nanotechnology, TechnologyReview.com and other media worldwide.
Developed a kinetic Monte Carlo approach for studying the dynamics of surface clusters.
Education
University of Warwick
Ph.D
Peking University
M.Sc
Peking University