Akademik Veri Yönetim Sistemi
2025 Interdisciplinary Conference on Electrics and Computer (INTCEC), North Dakota, Amerika Birleşik Devletleri, 15 - 16 Eylül 2025, ss.1234-1245, (Tam Metin Bildiri)
Agricultural producers frequently encounter difficulties accessing reliable and timely cultivation knowledge, constraining productivity and decision-making processes. We developed an agricultural assistant system based on RetrievalAugmented Generation (RAG) and Large Language Models (LLMs) to mitigate this challenge. A domain-specific dataset comprising approximately 19,000 question-answer pairs was constructed from Turkish Ministry of Agriculture guidelines, covering key topics such as pepper, tomato, cucumber cultivation, soil analysis, and pest and disease management. The dataset was structured into contextual Q&A format, and semantic embeddings were generated using OpenAI’s text-embedding-ada-002, subsequently indexed with FAISS (L2). The system retrieves the top three relevant fragments for each user query and generates formal answers through GPT-3.5Turbo. Experimental results indicate an average response time of 0.82 seconds with nearly complete contextual relevance. Both quantitative evaluations and pilot user feedback demonstrate that the proposed RAG+LLM framework is efficient, accurate, and scalable, providing real-time decision support for innovative agriculture applications.