Domain-Specific Generative AI and RAG Pipelines for Conversational Guidance in Experiential Learning Spaces: A Review and Proposed Conceptual Framework

The growing demand for autonomous, real-time interactive guidance in physical experiential learning environments has created a need for AI systems capable of delivering accurate, context-aware, and domain-specific conversational support. Physical experiential learning spaces—including interactive museums, science centers, and skill-training workshops—require guidance mechanisms that respond to learner queries with precision, spatial awareness, and pedagogical relevance. Standard Large Language Models (LLMs), however, are known to suffer from domain inaccuracy, lack of spatial awareness, and hallucination risk when answering domain-specific queries, which complicates their direct deployment in educational spaces. Three recurring limitations are particularly relevant: hallucination and factually incorrect generation; static and outdated knowledge that cannot reflect real-time changes in physical exhibits; and the absence of spatial grounding and domain-specific context awareness. To address these limitations, this paper proposes a domain-specific Generative AI architecture that combines Retrieval-Augmented Generation (RAG) with structured Knowledge Graphs (GraphRAG) and a multi-agent conversational pipeline for guidance in physical experiential learning spaces. The framework comprises three core components: a domain ontology and knowledge-graph construction module; a multi-agent pipeline integrating intent recognition, hybrid retrieval, and faithfulness-validation agents; and an interactive speech-enabled conversational interface. The GraphRAG component is designed to operate at two levels—local search (entity-centric, traversing direct neighbors in a knowledge graph) and global search (community-centric, aggregating over hierarchical clusters)—following the two-level GraphRAG design. The paper presents the proposed architecture and discusses the evaluation approach, operational challenges, and ethical considerations

کلیدواژه ها:

Generative AI ، Retrieval-Augmented Generation ، Knowledge Graphs ، GraphRAG ، Experiential Learning ، Multi-Agent Systems ، Digital Transformation