CS student @ HKUST · exploring how language models retrieve, reason, and reflect.
I'm interested in the messy part where retrieval meets reasoning — currently building agents on top of multimodal retrieval.
- FinDoc Agent — a vision-grounded RAG agent for financial documents, built on a self-finetuned ColPali
- Tinkering with LangGraph, post-training, and small-model inference
LLM agents · multimodal retrieval · post-training · inference efficiency
"What I cannot create, I do not understand." — Feynman