AI Retrieval System for Legacy Energy Asset Management
An innovative retrieval assistant enhances knowledge accessibility for traditional enterprise asset management systems within energy utilities. This solution covers three operational functions: Q&A for vendor documentation, Q&A for operational data store schemas, and Q&A for user interface usage. It integrates intent comprehension, query reformulation, hybrid semantic and vector retrieval, context engineering while adhering to token constraints, grounded answer generation, and deterministic hyperlink transformation. The data preparation pipeline focuses on semantic enrichment by incorporating descriptions for tables and fields. This research is available on arXiv under ID 2607.24792.
Key facts
- arXiv:2607.24792
- Energy utilities use long-lived enterprise asset management platforms
- Replacing platforms is cost prohibitive and operationally disruptive
- Retrieval assistant improves day-to-day knowledge access
- Three operational modes: vendor documentation Q&A, ODS schema Q&A, UI usage Q&A
- Runtime method includes intent understanding, query rewriting, hybrid retrieval, context engineering, grounded answer generation, hyperlink conversion
- Data preparation pipeline uses semantic enrichment with table and field descriptions
- Paper type: cross
Entities
Institutions
- arXiv