Researchers have published four papers addressing challenges in large language model (LLM) applications across different domains.
According to arxiv.org, a neurosymbolic architecture for enterprise AI agents introduces ontology-constrained neural reasoning to address hallucination and regulatory compliance issues. The system, implemented in the Foundation AgenticOS (FAOS) platform, uses a three-layer ontological framework comprising Role, Domain, and Interaction ontologies. In a controlled experiment with 600 runs across five industries including FinTech, Insurance, Healthcare, and Vietnamese Banking, ontology-coupled agents “significantly outperform ungrounded agents on Metric Accuracy (p < .001, W = .460), Regulatory Compliance (p = .003, W = .318), and Role Consistency (p < .001, W = .614).” The paper reports the production system serves 21 industry verticals with 650+ agents.
Separately, arxiv.org describes Alphacast, “an interaction-driven agentic reasoning framework” for time series forecasting using training-free LLMs. The system reformulates forecasting as a multi-stage workflow involving “context preparation, reasoning-based generation, and reflective evaluation,” and includes a toolkit with a feature set, knowledge base, case library, and contextual pool. Experiments across multiple benchmarks showed Alphacast “generally outperforms representative baselines.”
Additional papers address electromagnetic domain modeling through PReD, described as “the first foundation model for the EM domain,” and OmniFusion, which enables “speech-to-text, speech-and-image-to-text, and text-and-image-to-text translation” with a reported 1-second latency reduction in simultaneous speech translation.