Hi 👋, I am Yuan Tian. I am currently an Applied Researcher at 
Capital One AI Foundations,
doing research in AI for data.
I previously interned twice at
Adobe
as an Applied Scientist, working on the
AI
assistant for Adobe
Experience Platform.
I obtained my PhD in Computer Science
from 
Purdue University, advised by Professor Tianyi Zhang.
Along with my research at Purdue, I also worked on building a large-scale knowledge
graph for AI security, funded by NSF.
My research focuses on agentic data system, where I build coding and database agents for data-intensive tasksData-intensive tasks often involve overwhelming amounts of context (e.g., millions of data records in a database), which is challenging for agents to consume and process.. My work spans two complementary directions:
- Agent-centric: designing and scaling the interaction between AI agents and data environments via different approaches, such as post-training, attention steeringInspired by the anchoring effect, flexibly steering attention to the most relevant context, overcoming attention dilution, making generation more controllable and accurate., scaffoldingInspired by the grounding theory, scaffolding and decomposing the generation into mutiple accessible feedback loops (common ground), reducing task complexity, isolating errors, and improving performance., reasoning and verification.
- Data-centric: building data pipelines via programmatic data synthesis and augmentation for domain adaptation, as well as semantic enrichment to enhance the quality of data environments.
Data
