Leveraging Fine-Tuned Large Language Models for Multilingual Agricultural Chatbots to Improve Farmer Decision-Making

The agricultural sector faces challenges such as unpredictable weather, pest infestations, and inefficient resource management, impacting productivity and sustainability. This research presents an AI-driven chatbot using fine-tuned large language models (LLMs), including LLaMA 2 and Gemini-1.0-pro-002, to offer farmers accurate, context-aware, and multilingual support. By training on a dataset of 178,939 farmer queries and expert responses tailored to the Indian agricultural context, the models were fine-tuned using techniques like Parameter-Efficient Fine-Tuning (PEFT) and Quantized Low-Rank Adaptation (QLoRA). The Gemini-1.0-pro-002 model, demonstrating superior performance with 88% accuracy and a perplexity score of 18, was chosen for deployment due to its scalability and cost-effectiveness. The chatbot's multilingual capabilities and context-sensitive responses make it a valuable tool for assisting farmers with diverse linguistic backgrounds and agricultural needs. Its performance was validated through quantitative metrics, such as accuracy and response time, and qualitative expert feedback. Despite its effectiveness, limitations include restricted language support and the absence of real-time data integration. Future research will focus on expanding language capabilities, incorporating real-time IoT data, and broadening the dataset to cover more agricultural scenarios. This study establishes a foundation for AI-driven agricultural tools, promoting sustainable practices and informed decision-making.