Mou Sarkar , Pratham Prajapati , Rahul Dewangan , S Abhinav Raj and Sanjay Chatterji , Indian Institute of Information Technology Kalyani, India

This paper presents the design, implementation, and evaluation of a Soybean Crop Advisory Chatbot that utilizes Retrieval-Augmented Generation (RAG) techniques and Large Language Models (LLMs) to insight the farmers with precise, context-aware recommendations. The system integrates diverse data sources (research articles, extension bulletins, crop tables) into a unified knowledge base. Using semantic embedding and vector storage (via ChromaDB), the chatbot retrieves relevant information in response to user queries and formulates answers through a language model pipeline (LangChain) with prompt tuning for clarity and farmer-friendly language. Key challenges, such as extracting English content from bilingual PDFs, merging sentence fragments, choosing optimal text chunk sizes, and simplifying technical language for non-expert users, are addressed with custom processing strategies. We report development details including system architecture, data preprocessing, embedding generation, and prompt design. Sample queries and responses demonstrate the chatbots capabilities. Evaluation on test queries indicates high retrieval precision and user-friendly performance, suggesting the systems potential to improve soybean farming practices. The work discusses the limitations and future enhancements.

Soybean Crop Advisory, Chatbot, Retrieval-Augmented Generation (RAG), Large Language Models (LLMs), Semantic Embedding, ChromaDB, Data Preprocessing, Prompt Tuning, LangChain.