At Elephant Intelligence, I’m building a browser-use agent for web form filling and navigation. I built an evaluation benchmark and improved task success from ~20% to 100%, while making the agent ~2× faster and 25%+ cheaper than its closest competitor.
My work also spans 3D generative AI at Sceneverse and independent projects including a text-to-video pipeline, a RAG-based grading agent, and a VLM deepfake detector. I’m a co-first or first author on three papers covering agentic reinforcement learning, streaming vision-language models, and diffusion policy optimization.
