Zanwen Fu — ML Engineer & Founder

Zanwen Fu — ML Engineer & Founder Zanwen Fu Machine Learning Engineer (Agentic) at Robinhood · M.S. CS at Duke Agents that survive production. I build agent systems that hold up in production, not just in demos. The model is the easy part; the harness around it is where I spend my time. GitHub (opens in new tab)· LinkedIn (opens in new tab)· X (opens in new tab)· Email About My work lives in the layer between a capable model and a product people can actually rely on: orchestration, evaluation, rollback. The clearest proof is VYNN AI (opens in new tab), an agentic financial analyst I designed and built alone: a LangGraph multi-agent backend, a FastAPI orchestration layer, and a React dashboard, about 50K lines in all. It ran a ~500-user pilot and is now back online (opens in new tab), serving 5,000+ users. Before that I was a research engineer on AutoCodeRover (opens in new tab), an autonomous code-repair system whose core technology was acquired by Sonar; my pieces were the JetBrains plugin and the Self-Fix retry loop that pushed its SWE-bench Verified score to 51.6%. Alongside it I led research as first author on LUMINA (opens in new tab), a multi-agent framework for medical citation screening that reached 98.2% mean sensitivity across 15 systematic reviews. The thread through all of it is the harness, and that obsession now has a repository: taste (opens in new tab), an open-source operating system for agents that treats git as the memory substrate. I'm spending summer 2026 at Robinhood in Menlo Park, on the Agentic AI team. I'm open to full-time roles starting 2027. Duke University 2025 – 2027 M.S. Computer Science (AI/ML) · Graduate Teaching Assistant · scholar profile (opens in new tab) National University of Singapore 2021 – 2025 B.Comp. in Computer Science (Honours, Distinction) · Distinction in Software Engineering (opens in new tab)· exchange at HKU, Fall 2023 Experience Robinhood· Agentic AI Team current · May 2026 – Au