AI RESEARCH
SpecDB: LLM-Generated Customized Databases via Feature-Oriented Decomposition
arXiv CS.AI
•
ArXi:2605.31097v1 Announce Type: cross Mainstream relational databases ship a uniform feature set across deployments, although individual workloads exercise only a fraction of the available subsystems. We investigate whether a database can instead be generated on demand with a feature set matched to the target workload. We present SpecDB, a system that uses large language models (LLMs) to synthesize customized relational databases. We survey 9 production systems and decompose them into 10 functional modules, each further divided into implementation variants.