SAFAARI: Schema-Aware Framework for Accelerated Advertiser Response Intelligence
SAFAARI, a novel multi-agent framework, tackles the challenges of schema linking in NL-to-SQL systems designed for enterprise customer support. It employs dedicated agents for content, metadata, and orchestration. The performance of the system is assessed comprehensively using the SEAL metric. In tests involving five feature configurations, SAFAARI recorded a SEAL score of 81.66%, marking a 6.65% enhancement over the baseline, along with improvements in datapoint accuracy (5.51%) and schema-linking precision (4.69%). The research paper can be accessed on arXiv (2607.25042).
Key facts
- SAFAARI is a multi-agent framework for NL-to-SQL schema linking.
- SEAL is a composite metric for evaluating NL-to-SQL systems.
- SAFAARI achieved 81.66% SEAL score, 6.65% improvement over baseline.
- Datapoint accuracy improved by 5.51%.
- Schema-linking precision improved by 4.69%.
- The framework uses content, metadata, and orchestration agents.
- Five feature set configurations were tested.
- Paper available on arXiv: 2607.25042.
Entities
Institutions
- arXiv