Research to production, on purpose

I'm Huy-Dung — an applied AI engineer who spent three years in a PhD on multimodal perception and then went to ship it. I now work where computer vision meets agentic systems: real-time perception for a humanoid by day, conversational agents and tooling by night.

The thread across my work is taking a research-grade idea — a depth network, a tool-calling pattern, a clever loss — and turning it into something a non-research user actually relies on. I care about latency, about the failure modes nobody writes papers about, and about the unglamorous bits between a paper and a product.