RADIANT-LLM: AI Framework for Nuclear Safety Decision Support

A new framework called RADIANT-LLM (Retrieval-Augmented, Domain-Intelligent Agent for Nuclear Technologies using LLM) has been developed by researchers to improve decision-making in safety-critical nuclear engineering. This multi-modal retrieval-augmented generation framework tackles issues of fragmentation and hallucination found in pre-trained large language models through a local-first, model-agnostic design. It integrates a multi-modal document ingestion system with a structured, metadata-rich knowledge base, allowing for page- and figure-level retrieval from technical documents. Additionally, an agentic layer manages domain-specific tools, ensures responses are citation-backed with provenance tracking, and facilitates human-in-the-loop validation. Details of this research can be found in a paper available on arXiv (2604.22755).