Geospatial case study01
New Haven building-height spatial analysis
A 331,423-record spatial analysis with a focused interactive preview of 16,379 central New Haven polygons.
AI agent systems · data science
Reliable AI agents. Useful data systems.
I’m Xingye (Austin) Zhang, a Systems Engineering M.S. student at the University of Pennsylvania. I build and evaluate tool-using LLM agents, reproducible research pipelines, and data products grounded in evidence.
Agent reliabilityIntervention, recovery, evaluation
Applied dataGeospatial, statistical, ML
Systems thinkingTraceable, testable workflows
Selected work
Systems that can explain themselves.
I’m interested in work where rigor matters: agents that recover from blocked actions, maps that state their uncertainty, and workflows whose decisions can be traced back to evidence.
Current research02
Reliable tool-using agents
Evaluation design for runtime intervention, recoverable multi-step execution, leakage-aware traces, and closed-loop control.
Open-source extension03
Evidence-governed career operations
A human-in-the-loop workflow that separates evidence, approval, opportunity state, and irreversible submission actions.
Applied statistics04
From public data to honest conclusions
Course analyses spanning NYC rodent inspections and NBA salary prediction—including useful negative results and clearly stated limitations.
How I work
Evidence first, then action.
Strong systems are not defined only by a model or interface. They need clear boundaries, reproducible checks, and an honest account of what the evidence supports.
Make the state visible
Structure inputs, decisions, and traces so a result can be inspected instead of merely trusted.
Test the failure path
Evaluate blocked actions, recovery, stale data, and edge cases—not only the clean happy path.
State the boundary
Separate measured outcomes from inference, archived results from reruns, and prototypes from production systems.
Experience
Research meets engineering.
My path combines statistics and economics with systems engineering, agent evaluation, backend development, and geospatial research.
AI Agent Systems Research Collaboration with UConn Researchers
Research and engineering for reliable tool-using agents, runtime intervention, evaluation, and closed-loop recovery.
University of Pennsylvania
M.S. student in Systems Engineering, extending a foundation in statistics and economics into AI and systems work.
University of Connecticut
Research Assistant working with satellite imagery, spatial statistics, building-height data, and HPC workflows.
GBCS — SkyIT Services
Backend Developer Intern contributing to the delivery and release of the Voop application for a client, REST APIs, and a Firebase-to-MySQL migration.