Yixuan (Ethan) Chen, Applied AI undergraduate at the University of Hong Kong. Available for internships from December 2026.
FirmGround, my own product
A research agent for document-heavy work. It searches standards, filings and textbooks, runs any arithmetic in code, and cites the passage behind each claim. 81.33% on FinanceBench, against 21.33% for naive RAG.
Read the case studyWYNI Technology, internship
Minutes from Cantonese–English meetings, and claim emails turned into review-ready workbooks.
WYNI Technology, internship
The agent sees three meta-tools and loads a tool’s schema only when it needs it.
FirmGround, my own product
Reconciliation where algorithms do the matching and accountants approve what agents suggest.
HKU Visual AI Lab, research
Screening reports written from detector output, with a confidence score on every claim.
Before I trust a component, I test it on a set built to break it. These three results decided what I shipped.
X-ray screening reports need a verifier that catches claims the evidence doesn’t support. I chose it on a benchmark I built, not on a leaderboard.
200 simulated reports, 921 claims, 20 types of planted error.
X-ray screening reportsSame model and embeddings as a naive RAG baseline. The only change is the agent loop and its retrieval tools, so the gap is theirs.
FinanceBench open set, 150 questions over public company filings.
FirmGround Agentic RAGHiding tool schemas saves tokens only if the agent still picks the right tool, so two thirds of the cases were written to be confusable.
confusable case (40) standard case (20)
60 tasks across 83 tools.
One gateway for 83 toolsI’m Yixuan Chen, and I go by Ethan. I study Applied AI at the University of Hong Kong. Most of what I’ve built started from someone else’s slow task: claims handlers preparing case files, accountants reconciling statements, a school searching its own textbooks. I like the part where a tool leaves the demo and becomes something people keep using.
I’m a student adviser for the Applied AI programme and present at student-led GenAI seminars.