Open to opportunities

Yuying Liu

@yuyingliu

Message

Applied scientist specializing in ML, AI agents, and data-driven dynamical systems research.

United States

Message

What I'm looking for

I seek roles building production ML systems or research-driven AI solutions that balance scalability, privacy, and interpretability, with opportunities for technical leadership and impactful products.

I am an applied scientist with a PhD in Applied Mathematics and extensive experience building machine-learning solutions for real-world systems, including AI task agents, fraud detection, federated learning, and forecasting models. At Amazon I designed ReAct agents integrated with Playwright, fine-tuned compact language models, and engineered MCP integrations to accelerate backend execution, delivering measurable runtime and task-completion improvements.

My research background includes physics-informed neural networks, Koopman-based autoencoders, and high-performance algorithms for dynamical systems, with publications, patents in progress, and open-source contributions such as PyNumDiff and PySINDy. I bring cross-domain expertise from industry and research labs to drive production-ready ML systems that balance interpretability, privacy, and scalability.

Experience

Work history, roles, and key accomplishments

Current

Applied Scientist II

Current

Amazon

Mar 2022 - Present (3 years 8 months)

Designed and built AI task agents for the Amazon Advertiser Console, fine-tuned compact language models to cut runtime 42% with minimal accuracy loss, and integrated backend MCP servers to achieve 12× speedup and 15% higher task completion; developed ML solutions for fraud detection and a privacy-preserving federated learning architecture.

Playwright AWS Sagemaker EC2 S3 VPC Federated Learning Fraud Detection Tensor Models

AI Resident

Google X

Jan 2022 - Mar 2022 (2 months)

Worked on sensor placement and signal processing in the Chorus team, applying CNN and Bayesian methods to improve indoor localization accuracy by 6%.

Signal Processing CNN

Research Scientist

Mitsubishi Electric Research Laboratories

Sep 2021 - Dec 2021 (3 months)

Proposed a physics-informed autoencoder architecture leveraging Koopman theory to generate interpretable latent representations for nonlinear dynamical systems and conducted data-driven HVAC control research (US patent filed).

Dynamical Systems

Data Scientist

Bill & Melinda Gates Foundation

Jun 2020 - Sep 2020 (3 months)

Predicted high-resolution economic indicators for Nigeria and Pakistan using Bayesian models and transfer learning on satellite imagery, improving model accuracy by 2.8%.

Transfer Learning Remote Sensing Geospatial analysis

Computational Mathematician

Lawrence Livermore National Lab

Jun 2019 - Sep 2019 (3 months)

Developed adaptive convolutional autoencoders for multi-resolution compression of global sea surface temperature data and implemented C++/MPI parallel algorithms for high-fidelity simulation on Sierra.

Convolutional Autoencoders Multi Resolution Compression C++MPI High Performance Computing

Education

Degrees, certifications, and relevant coursework

University of Washington

Doctor of Philosophy, Applied Mathematics

2017 - 2022

Grade: 3.8/4.0

Activities and societies: Thesis advisors: Nathan Kutz and Steve Brunton; Boeing Fellowship (2017-2019).

Completed a PhD in Applied Mathematics with thesis on neural networks for nonlinear dynamical systems, focusing on modeling and compressible representations for high-dimensional systems.