I study Information and Computing Science at Xiamen University.
My interests sit where mathematical modeling, generative AI, and dependable software meet.
Selected work / Open source / Current focus / Beyond the model
| AI for Science Physics-aware generation, mathematical modeling, and optimization. |
Reliable AI systems Reproducibility, validation, failure handling, and security boundaries. |
| Product engineering Browser experiences, typed frontends, APIs, and cloud data paths. |
Open-source quality Focused fixes, regression tests, and reviewable design decisions. |
| LEARNING SYSTEMS / RELIABILITY A browser-based Python learning experience built around short lessons, real code execution, and observable progress. My contribution focused on dependable evaluation and the engineering quality of the public learning path. Visit the public application |
| InvokeAI SECURITY / PYTHON Closed two SSRF paths in external image downloads and injected HTTP sessions, backed by adversarial socket-level tests (#9525, #9524). |
| Diffusers GENERATIVE AI / PYTHON Corrected required-input metadata for custom Mellon pipeline blocks and added focused regression coverage (#13888). |
| sktime ML INFRA / TYPING Migrated the multithreading capability to the typed tag registry while preserving lookup behavior (#10848). |
| Soda Core DATA / PRIVACY Prevented SQL Server connection parameter values from leaking into logs (#2762). |
| freeCodeCamp AST / JAVASCRIPT Replaced syntax-specific validation with AST inspection so declarations, expressions, and arrow functions follow the same curriculum rule (#69520). |
I am exploring physics-aware generative modeling and optimization for AI4S, alongside reliability and security in generative AI infrastructure. I prefer work that can be reproduced, tested under failure, and explained from a design decision to an observable result.
I also care about visual explanation, clean interfaces, and concise technical writing. Research is easier to trust when another person can understand the assumptions, run the code, and see how it fails.
Open to thoughtful collaboration around AI4S, generative systems, and reliable ML infrastructure.