Khang Thai

I'm a second-year master's student in Applied Data Science at the University of Southern California, and I finished my B.S. in Statistics and Data Science at UCLA in 2025. My training is classical — regression, model diagnostics, cross-validation, knowing when a result is real — and I've spent the last year pushing that foundation into applied AI.

At Hologic I architected a hybrid retrieval pipeline — dense vector search, BM25 and cross-encoder reranking over 15,000+ chunks from 270+ controlled assay documents, which doubled top-1 retrieval accuracy — and now build the multi-step LangGraph agents that sit on top of it, with tool calling and stateful human approval gates. Those workflows reach GitHub Copilot over the Model Context Protocol, and an evaluation harness of 195 checks covers schema, data fidelity and reproducibility before a draft goes anywhere. In a regulated setting a confident wrong answer is worse than no answer, so most of the engineering went into keeping model output out of signed deliverables and making the system say what it actually knows.