Authors: Rohan Mehta, Arvind Sethi, Nisha Kulkarni, Vasudev Sharma
Abstract: Enterprises operating in complex digital ecosystems face accelerating growth in data volume, reporting demands, and governance obligations. Traditional rule-based automation remains insufficient for interpreting ambiguous business semantics, harmonizing heterogeneous information assets, or sustaining consistent reporting logic across distributed platforms. This study introduces a large language model powered semantic automation engine designed to unify enterprise reporting, knowledge extraction, and end-to-end data lifecycle governance. The research focuses on the central challenge of operationalizing generative models, retrieval-augmented reasoning, and dynamic semantic alignment to automate high-stakes analytical and compliance workflows while maintaining auditability, accuracy, and policy adherence. Using a mixed methodological approach that combines empirical prototyping, workflow instrumentation, and qualitative validation with enterprise architects, the study develops a layered architecture integrating semantic parsers, governance ontologies, vector-indexed knowledge repositories, and automated lineage reasoning. Findings show that LLM-driven inference strengthens metadata completeness, reduces manual reconciliation cycles, enhances cross-system reporting consistency, and improves lifecycle visibility from ingestion to archival. The study contributes a scalable framework for semantic automation, a reference ontology for enterprise reporting logic, and a set of design principles supporting trustworthy, context-aware automation across data-intensive environments.
DOI: http://doi.org/10.5281/zenodo.17895633
