Tiancheng Xu
Contact
San Francisco, CA
Email: tianchengnxu[at]gmail[dot]com
LinkedIn · X
About
I work on research strategy at ByteDance Seed. My work covers: recursive self-improvement; how frontier capability becomes enterprise revenue that funds the next model; the training data and RL environments behind long-horizon work and, eventually, autonomous research; and the culture and organization a long-term research mission needs.
Before that I was one of Doubao's first product leads, based in Silicon Valley, building the end-user experience for what is now Doubao Work (docs, storage, browser, front-end coding) as the app grew from launch to 300 million monthly users. I have an M.S. from UC Berkeley and a B.S. from the University of Washington.
Selected work
- Contextual assistance at Doubao. Work on “Ask Doubao”: bringing the right context into a conversation, investigating failure cases, and evaluating whether an answer actually helps.
- Documents as a workspace. Product work across Doubao’s AI reading and cloud-drive experiences, including source citations, document navigation, and ways to use files as context.
- AI for working with code. Product exploration for Doubao × MarsCode: helping developers understand unfamiliar repositories and move between reading, building, and troubleshooting.
- AI data and model strategy. Research into training-data suppliers, reinforcement-learning environments, evaluation, and the domains where better AI capabilities could matter most.