I build applied AI systems that connect data, models, and real operating workflows. My background combines graduate quantitative training, end-to-end machine-learning projects, and hands-on work translating ambiguous business problems into testable, reviewable AI workflows.

• Applied Machine Learning Portfolio — computer vision, NLP, and explainable tabular modeling with model comparisons, verified saved metrics, limitations, and explicit team attribution.
• AI-Native Foreign Trade Operations Workspace — a project I am currently developing around real operating needs, with the goal of connecting channels such as Alibaba.com, independent websites, email, and social networks into a reusable AI-driven workflow for lead development, outreach, CRM, follow-up, and operational learning. The public repository is a sanitized demo version.

• End-to-end machine-learning evaluation and error analysis
• LLM applications, structured outputs, tool use, and human-in-the-loop controls
• Reusable data and workflow pipelines for business operations
• Agent-oriented product thinking: task decomposition, review gates, auditability, and continuous improvement
• Technical communication across engineering, data, product, and commercial teams

• Lyslle — an early Streamlit safety-classification prototype with explicit limitations and human-review routing. It is presented as a small prototype, not a deployed clinical system.

I care about measurable outcomes, honest evaluation, clear ownership, and AI systems that leave behind reusable workflows rather than one-off outputs.