Reproduce “Ontology-Constrained Neural Reasoning in Enterprise Agentic Systems: A Neurosymbolic Architecture for Domain-Grounded AI Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Thanh Luong Tuan, Abhijit Sanyal
This research introduces a neurosymbolic architecture for enterprise AI agents, integrating a formal three-layer ontology (Role, Domain, Interaction) with Large Language Models. It empirically demonstrates improved performance in domain-specific reasoning, regulatory compliance, and role consistency, especially in domains with sparse LLM training data and for enterprise-specific metrics.
Key insights
- The "Inverse Parametric Knowledge Effect" (Inverse PKE) indicates that ontological grounding provides the greatest value where an LLM's pre-existing parametric knowledge is weakest, such as for highly specialized enterprise metrics or less represented languages/domains.
- Formal enterprise ontologies significantly enhance an agent's ability to achieve high Metric Accuracy and Role Consistency by providing structured definitions for domain-specific KPIs and decision patterns.
- While RAG with curated chunks can perform comparably for lexical recall, the structural and relational grounding provided by formal ontologies offers a distinct advantage for complex reasoning tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.