SpecDB: LLM-Generated Customized Databases via Feature-Oriented Decomposition
SpecDB is a system that leverages large language models (LLMs) to create tailored relational databases designed for specific workloads. Researchers examined nine production systems, breaking them down into ten functional modules, each featuring various implementation options. They utilized the FODA feature model, enhancing it with a cooperative edge to represent dependencies across modules, resulting in a dependency graph known as DBGraph. SpecDB implements DBGraph through a multi-layered module-construction pipeline, where each module is produced, validated, and integrated by a specialized subagent, which is guided by three internal agents: Main. This research is available on arXiv with the identifier 2605.31097.
Key facts
- SpecDB uses LLMs to generate customized relational databases.
- 9 production systems were surveyed and decomposed into 10 functional modules.
- FODA feature model extended with cooperate edge to create DBGraph.
- DBGraph captures cross-module dependencies.
- SpecDB uses a layered module-construction pipeline.
- Each module is generated, validated, and integrated by a dedicated subagent.
- Subagents are driven by three inner agents: Main.
- Published on arXiv with ID 2605.31097.
Entities
Institutions
- arXiv