A POSTECH research team led by Professor Wook-Shin Han of the Department of Computer Science and Engineering and the Graduate School of Artificial Intelligence, along with Ph.D. candidates Taesung Lee and Jaehyun Ha, has developed TurboLynx, an engine capable of analyzing complex, highly interconnected data up to 184 times faster than existing systems. The findings are scheduled to be presented at VLDB 2026, a top-tier international conference in the database field.
The same underlying technology is at work when Netflix recommends content tailored to individual users' preferences, financial institutions detect suspicious transactions, and generative AI understands connections between people and concepts—the ability to interpret relationships. Data that captures complex networks of connections—among people, between products and transactions, or among words—is known as graph data. Graph database technologies store and analyze these densely interconnected webs of information, much like a spider's web.
The challenge lies in the diversity and irregularity of real-world data. In enterprise environments, data structures change frequently, and new attributes can be added at any time. Such "schemaless" data, whose format is not fixed in advance, offers flexibility, but existing systems often slow down significantly when handling analytical tasks such as aggregation and statistical analysis. It is like a restaurant where every customer submits a differently formatted order, forcing staff to manually read through each one just to tally the day's sales.
To address this challenge, the research team redesigned the entire system, from data storage to query processing and query optimization. The key idea was to group similar data together and process them collectively.
The newly developed engine, TurboLynx, automatically groups data with similar characteristics and stores each group in a columnar format optimized for analytics. This eliminates the need for the system to interpret the structure of the data every time it is read, while significantly reducing unnecessary memory usage. TurboLynx also reduces the excessive intermediate-result explosion during complex multi-step traversals and enables analytical queries to run more efficiently.
TurboLynx's performance gains were demonstrated quantitatively through standard benchmark evaluations. In internationally recognized benchmark evaluations, the engine was about 184 times faster than existing graph database systems and up to 41 times faster than relational database systems. On a large-scale Wikipedia-based knowledge graph dataset, TurboLynx outperformed the best-performing competing system by approximately 19 times, demonstrating its potential for real-world industrial applications.
As the use of complex relationship data rapidly expands across areas such as generative AI, recommender systems, financial security, and biomedical data analysis, this technology could lay the foundation for near-real-time data analysis. Just as a detective in a film can instantly trace a vast criminal network, companies will be able to much more quickly interpret and leverage the connections embedded in massive datasets.
Professor Han said, "We expect TurboLynx to help companies make broader use of complex graph data in real-world analytics and services," adding, "We plan to continue our research so that the system can support real-time transaction processing and serve as long-term memory for AI agents."
TurboLynx supports Cypher, an industry-standard graph query language, and allows general users to interact with the system using natural-language queries. The system has also been released as open source, with related information and access available through the project website at https://turbolynx.io .
This work was supported by an Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea Government (MSIT) under the project "Development of a Distributed Graph DBMS for Intelligent Processing of Big Graphs," and by a National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT). The natural-language query translation research was supported by the IITP SW StarLab project "Development of a Conversational, Self-Tuning DBMS," funded by the Korea Government (MSIT).
