We build data systems that learn.
builds data systems for the AI era — towards systems that are principled, grounded in theory and quantifying how far we can improve or go wrong; adaptive, learning from execution rather than predetermining everything; and declarative, so users say what they want, not how to get it. As a long-term goal, we aim to build a reliable AI-driven research assistant for reducing obvious/manual tasks so we can focus on genuinely interesting problems.
We are recruiting graduate students and undergraduate interns. Learn more
Four lenses into data systems
Our work can be viewed by domain, by system layer, by methodology, or by the dimensions of heterogeneity we tackle. Click any item to see what it means and the papers behind it.
Pure DB
Core database technology (mainly OLAP): adaptive/declarative query processing, hardware-aware query processing, physical operator design, and query optimization.
AI4DB
AI inside the database, learned components that replace heuristics and manual labor inside data systems: query optimization and NL-to-SQL translation.
DB4AI
Data systems for AI, the infrastructure AI workloads run on: LLM inference serving, vector query processing, semantic operators and engines over unstructured data, and datalakes.
Three domains covering the DB–AI landscape.
Query Interface
how you askQuery Optimization
how it plansQuery Processing
how it runs over dataData Maintenance
how data livesThe stack of a data system, top to bottom.
Our methodologies overlap — the interesting problems live in the intersections.
Click a line for the overview, or a point for its papers.
Click a point above (e.g. Relation, GPU) to see the papers that use it.
Related papers
Working with us
We look for students who are curious about how data systems actually work — and restless about how they should work in the AI era. Graduates, and undergraduate interns who want 3-6 months of research experience are also welcome. A strong background in any one of systems, algorithms, theory, mathematics, or machine learning is a great starting point.
Beyond background, we look for students who enjoy the process of solving hard problems with the grit to stay on them, speak up with their own ideas while staying open to different ones, enjoy growing and keep asking for one more step, and just start without worrying too much.
As an early member of a new lab you will work directly with the professor, own your research agenda, and aim for top database venues such as SIGMOD, VLDB, and ICDE.
If interested, please send an email with your CV, transcript, and a short note on what you would like to work on.
We are located in Room 4405, Building E3-1 (School of Computing), KAIST, Daejeon, South Korea.