Come and Build a Silicon Dream with Me
Embedded Files
About me
About me
👋 Hi, I’m Yingchao Li, a research engineer at Anthropic and a computational finance graduate from the University of Washington. My work centers on large language models, multi-agent systems, and the design of human-centered AI. I’ve contributed to projects like Claude Series and open-source initiatives such as Qwen, optimizing both performance and alignment across text, code, and multimodal data.
In parallel with industry, I conduct research on AI-assisted decision-making, agentic workflows, and sign language animation systems—bridging machine learning, optimization, and accessibility. I’m passionate about building intelligent systems that reason, adapt, and interact meaningfully with people.
Education
University of Washington - Seattle
University of Washington - Seattle
MS in Computational Finance and Risk Management, Sep 2022 - March 2024
Georgia Institute of Technology
Georgia Institute of Technology
MS in Computer Science, Sep 2024 - Present
Academic Experience
I am currently serving as a research assistant at Foster School of Business and Department of Electronic and Computer Engineering at University of Washington, instructed by Prof. Leonard Boussioux and Prof. Banghua Zhu.
Research Interest
Multi‑Agent Systems, MAS
Agentic LLM Architectures
Multimodal HCI and Accessibility AI
Profession Experience
Anthropic PBC, Seattle, WA / New York City, NY
Anthropic PBC, Seattle, WA / New York City, NY
Research Engineer, Tokens (Pre-training) | May 2025 – Present
Claude 4.0 & Circuit Tracing Tools:
Optimized constitutional AI training data and RLHF pipelines, enhancing model alignment and safety.
Developed scalable frameworks for automated data filtering, content quality assessment, and bias detection.
Designed multimodal tokenization systems for processing text, code, and structured data, focusing on vocabulary optimization and cross-domain token representation.
Machine Learning Engineer, RL Engineering | Jan 2024 – Apr 2025, Seattle, WA
Claude Series (Claude 3.5, 3.7, and 4.0):
Contributed to the development of the Model Context Protocol (MCP), improving model interactions with external tools and data sources.
Built Claude’s agent system, enabling autonomous execution of complex, open-ended tasks such as code generation and problem-solving through reinforcement learning and multi-agent interaction techniques.
Alibaba Cloud, Qwen-LM Team, Hangzhou, China
Alibaba Cloud, Qwen-LM Team, Hangzhou, China
Algorithm Engineer – Multimodal | Mar 2023 – Sep 2023
Qwen Open-Source Large Language Model:
Reduced model training time by 32.6% using mixed precision training techniques, facilitating training on larger-scale datasets.
Enhanced performance and developed documentation for Qwen’s open-source release, promoting its adoption and usage within the broader community.
High-Flyer AI, Hangzhou, China
High-Flyer AI, Hangzhou, China
Machine Learning Engineer | Sep 2021 – Aug 2022
DeepSeek & OpenCastKit Projects:
Improved GPU cluster software and resource allocation for the Fire-Flyer Supercomputer Cluster, notably optimizing GPU collaboration without NV-Link.
Co-developed OpenCastKit, integrating FourCastNet and GraphCast models, leveraging parallel communication optimizations for improved numerical prediction capabilities.
SenseTime Technology, Shanghai, China
SenseTime Technology, Shanghai, China
Data Mining Engineer | Jun 2021 – Sep 2021
Autonomous Driving Data Pipeline:
Developed advanced data mining algorithms to identify and handle long-tail scenarios (extreme weather, complex traffic behaviors, rare obstacles).
Constructed an automated data mining pipeline integrating multimodal data (point clouds, images, text), significantly enhancing data labeling efficiency and annotation quality.
Page updated
Google Sites
Report abuse