AI-Powered Agritourism Chatbot: A Domain-Adapted RAG Chatbot with Transformer NLU for Rural Business Decision Support

Xue A. Dong

Keywords: retrieval-augmented generation (RAG), transformer NLU, intent classification, policy compliance, rural decision support, agritourism

Abstract: I present the AI-Powered Agritourism Chatbot, a decision-support system that helps farmers and ranchers navigate planning, marketing, risk, and compliance. Stage 1 (summer 2025) delivered a working RAG assistant connected to curated regulations, guidance, and case materials, with tools for business planning, marketing, risk management, compliance, and impact assessment. Stage 2 adds a domain-tuned transformer NLU (intent classification and slot tagging), a multi-hop retrieval controller, and a safety/policy layer that filters claims before responding. We evaluate with offline metrics (intent accuracy, slot F1, EM/F1, toxicity/grounding) and field tests in Extension settings (task success, latency, user satisfaction), including A/B comparisons to the Stage-1 baseline. The approach, domain-specific NLU, safety-aware RAG, and task-centered evaluation, targets underserved rural communities and transfers to other Extension contexts such as small-business advising and food systems.

Submission Number: 277