This study presents the design and implementation of an agricultural advisory chatbot to help farmers access reliable information on crop varieties and cultivation techniques, with a focus on rice and mango. The proposed system integrates a large language model (Gemini 2.5 Pro) with a Retrieval-Augmented Generation architecture to mitigate hallucination and improve factual accuracy in domain-specific responses. The knowledge base is constructed from authoritative agricultural technical documents, including crop variety descriptions, cultivation procedures, and pest and disease management guidelines. These materials are systematically preprocessed, segmented, and indexed in the Qdrant vector database to enable efficient semantic retrieval within the Retrieval-Augmented Generation pipeline. To enhance retrieval robustness, the system employs a hybrid search strategy that combines keyword-based retrieval and dense vector search, followed by a Cross-Encoder re-ranking module to optimize contextual relevance before response generation. System performance is evaluated using the Retrieval Augmented Generation Assessment framework. Experimental results demonstrate high reliability, with a Faithfulness score of 91.43% and an Answer Relevancy score of 95.52%. The findings indicate that the proposed approach can deliver accurate, context-aware, and practically applicable agricultural recommendations, highlighting its potential to support digital transformation and AI-driven solutions in smart and sustainable agriculture.