Research · Systems · Safety
Publishes as Shixuan Lin
AI Safety Researcher
Building Safe Tool-Using LLM Agents
Current work examines failure modes in tool-using LLM agents and translates technical findings into operational controls and digital workflows.
Professional record
Experience
Honghuayuan FabricQingdao, China
Leads development of a B2B digital platform and designs workflow and data architecture across inventory, procurement, production, sales, and finance.
Academic client project
GreenworksNew York, NY, U.S.A.
Analyzed outdoor power-equipment markets and developed recommendations for segmentation, positioning, pricing, promotion, and sales-channel strategy.
NYU Real World Client Project — an academic client project, not employment by Greenworks.
Experience
HMD Global U.S.A.San Diego, CA, U.S.A.
Conducted market and product analysis for Nokia mobile devices and used Python and SQL to prepare and analyze sales, market, web, and customer-feedback data.
Experience
Shanghai Longyi Enterprise Management GroupShanghai, China
Analyzed China's sugar-free food and beverage industry, estimated market size, and contributed to a 9,930-word consulting report.
Experience
Huaxia Bank, Qingdao BranchQingdao, China
Analyzed branch-issued bonds and collected financial data for three institutional clients, while supporting internal review and reporting documentation.
Academic background
Education
New York UniversityConcentration in Database Technologies · GPA 3.90/4.00New York, NY, U.S.A.
Selected coursework included Database Design and Management, Information Technology and Data, Information Security Management, and Quantitative Models for Decision-Making.
Education
Macau University of Science and TechnologyGPA 3.42/4.00 · WES GPA 3.70/4.00Macau SAR, China
Selected coursework included Corporate Finance, Asset Pricing, Financial Derivatives, Financial Econometrics, Banking Management, and Operations Management.
Education
Fudan UniversityShanghai, China
Selected coursework: Python Programming and Econometrics.
Peer-reviewed work
Publication
International Conference on Informatics Engineering and Information Science · Springer Nature SingaporeCo-author and Empirical Researcherpp. 152–161
Liang, W., Pan, R., Cao, Y., Wang, W., Lin, S., and Zhao, Y. DOI: 10.1007/978-981-99-3618-2_15.
View DOI recordSelected research & systems work
A four-layer framework mapping public AI risk guidance to testing requirements, evidence structures, deployment decisions, and mitigation verification.
A multilingual NLP system combining GPT-4 and a fine-tuned BERT model for structured review of xenophobia, toxicity, misinformation, and biased framing.
UNICC-sponsored NYU applied project — not employment by UNICC and not a claim of deployment or institutional endorsement.An end-to-end product and operations system spanning a WeChat Mini Program, back-office workflows, data architecture, deployment, and adoption.
Employment project completed in the role of Product and Operations Manager at Honghuayuan Fabric.Methods & tools
LLM evaluation · Threat modelling · Adversarial testing · Risk taxonomies
GPT-4 · BERT · Text classification · Multilingual content safety
Python · SQL · PostgreSQL · TypeScript