(C:)
profile
AgentProvenance
PhD Student · TU Graz · Visualization Design Lab
Yifan (Eden) Wu
I study how integration, visualization, and provenance can make complex data and AI workflows understandable and trustworthy.
I am a PhD student at TU Graz, advised by Prof. Alexander Lex in the Visualization Design Lab. My research connects data visualization and visual analytics (VAST), provenance, and data integration, with a particular interest in biomedical data. Previously, I worked with Prof. Juliana Freire at NYU VIDA on data discovery and integration.
01
Research direction
From heterogeneous data to visual understanding and accountable computation, with LLMs supporting each layer rather than forming a separate research silo.
Integration
Discover, align, and harmonize heterogeneous data. From heuristics to embedding-based approaches, then LLMs.
Data integration · Schema matching · Biomedical dataVisualization
Build interactive views that help experts explore complex data and do decision-making. LLMs assist and coevolve with human experts.
Visual analytics (VAST) · Human–AI sensemakingProvenance
Capture the data, decisions, and dependencies behind system and agent workflows so they can be inspected, reproduced, and reused.
Provenance · Agentic systems · Reproducibility02
Publications
Peer-reviewed work and current manuscripts. Eden Wu is highlighted in each author list.
LakeQA: An Exploratory QA Benchmark over a Million-Scale Data Lake
A benchmark for agents that must search and reason over a 9.5 TB heterogeneous data lake.
InsightAR: A Tool for Multi-modal Summarization and Interactive Analysis of AR-based Egocentric Task Videos
An interactive system for summarizing and analyzing multimodal egocentric task videos.
VizAgentBench: Benchmarking Multimodal Agent Reasoning on Coordinated Multi-View Visual Analytics Tasks
A benchmark for multimodal agents that interact with coordinated, multi-view visual analytics dashboards.
Large Language Models for Data Discovery and Integration: Challenges and Opportunities
A research agenda for applying language models to semantic heterogeneity in data discovery and integration.
Enhancing Biomedical Schema Matching with LLM-Based Training Data Generation
Uses LLM-generated training pairs and contrastive learning to improve in-domain biomedical schema matching.
03
Selected systems
Research ideas made testable through working software.
Provenance · Visual analytics · Lead author
AgentProvenance
The open-source prototype behind AgentTrails transforms agent traces into provenance graphs for inspection, comparison, and reuse.
Join discovery · Semantic sketches · Second author
MosaicJoin
Compact semantic sketches make value-level join discovery practical for large and heterogeneous columns.
Spatial data · LLM assistance · Second author
UrbanTrace
A visual, node-based environment for discovering, integrating, and analyzing urban spatial data.
Visual analytics · LLMs
BDIViz
Interactive schema matching and benchmarking for experts working with biomedical data.
Data integration · Language models
Magneto
A cost-aware retrieval and reranking architecture for accurate schema matching.
AutoML · Reinforcement learning
Alpha-AutoML
Pipeline search and reproducible model development across multiple data modalities.
04
Experience
A research path grounded in production engineering.
PhD Student · TU Graz
Working with Prof. Alexander Lex in the Visualization Design Lab on provenance, reproducibility, and visual analytics for agentic systems.
Research Engineer · NYU VIDA
Worked with Prof. Juliana Freire on research systems spanning schema matching, visual analytics, data-lake question answering, and AutoML.
Startup technical leadership
Led product and engineering work at RichCRM and Arkive; co-founded TagMe Network. Shipped AI-assisted and full-stack products from prototype to pilot.
ICT Software Developer · Huawei
Built packet-core networking and high-availability container infrastructure in Go and C; received the Huawei Future Star Award.
05
Beyond the lab
I enjoy cooking, hiking, and camping—especially cooking while camping. I keep a small culinary archive here as a record of meals, experiments, and good times shared outdoors.
Open culinary archive