Benchmark RadarDoc2DB-Bench
Doc2DB-Bench evaluates document-to-database construction, converting long heterogeneous documents into normalized relational databases with entity identities, keys, cross-table links, and integrity constraints. It includes 203 document instances across 42 schemas and 7 domains, with fine-grained capability annotations for intra-table extraction and inter-table reasoning.
- Released
- 2026-08-09
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Existing document-to-table benchmarks overlook relational database requirements such as normalization, entity resolution, and cross-table consistency. Doc2DB-Bench addresses this gap by providing a testbed for assessing LLM-based systems in realistic database construction, supporting analytics, compliance, and decision-making applications.
Motivation
Practical AI systems increasingly need to turn long, heterogeneous documents into queryable relational databases, not isolated spreadsheets.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.