Multilingual LLM evaluation harness for query classification, summarisation, RAG, and PII leakage under compositional task interference. Accepted to 3 NeurIPS 2026 workshops.
Troy Hanfei Tian
I am a second-year undergraduate student at the University of California, Los Angeles, concentrating in Computer Science, Asian American Studies, and Real Estate. My work and interests primarily lie in AI safety, including technical research, policy/governance, and especially their intersection. I also write on culture, politics, music, social justice, and other topics.
Pinned Projects
Referenceless evaluation of VLM-generated image descriptions for accessibility, comparing judgments from blind and low-vision and sighted users.
Inglewood prototype for exploring local businesses and neighbourhood support needs through housing data and reviewed street-level imagery.
Original Civilization VI mod adding Attila as leader of the Huns, built for import into and development in the Firaxis ModBuddy IDE.