I'm Jiaxin Huang, a mathematics student at Sun Yat-sen University. My work sits in applied AI: how intelligent systems can be evaluated, adapted, and deployed when the data is messy, the constraints are local, and the cost of mistakes is real.
I currently focus on NLP in the context of China's non-performing asset industry. I'm building TrustNPA-Bench, developing the TrustNPA due diligence platform, and exploring zkML for verifiable, privacy-preserving model inference.
I'm also building AI Hainan (海南AI创新应用实验室), an applied AI lab for moving frontier ideas into practical regional deployments. I host Off-Shore, a podcast about the people and ideas shaping Hainan's Free Trade Port.
For me, the hardest and most interesting AI problems are no longer only about the model. They show up when systems have to work in the world.
Working Paper / Benchmark
A layered benchmark for testing how LLMs read and reason over Chinese NPA legal documents, built around noisy documents, domain constraints, and deployment risk.
Research Note
A research note on using zero-knowledge machine learning to make model inference verifiable and privacy-preserving in sensitive due-diligence settings.
Full-stack Platform / Research Testbed
A full-stack due diligence platform for Chinese NPA workflows, used as both product infrastructure and a research testbed for document understanding, risk analysis, and evaluation.
A short plog on change, uncertainty, self-correction, and taking action.
A plog on matrix multiplication, the brain, information structure, and intelligence.
A plog on belief, introspection, interpretation, and inner change.