KrishiAI: architecture for harnessing capabilities of LLMs for delivery of accurate cultivation information to farmers

With the constantly evolving agricultural environment, farmers need timely information relevant to their context to make informed decisions. Although several digital solutions have been introduced to address this requirement, these disregard farmers' context in providing information, including information quality. To address this gap, we developed a chatbot in this study incorporating the latest advances in computing techniques. We created an architecture to harness current Large Language Model (LLM) capabilities, including knowledge modeling, to deliver context-specific information to users through questions and answers. The context was established from previous interactions in the form of a dialogue. Neo4J graph database was used to create the knowledge base, ensuring efficient storage and retrieval of contextual information. LLM was used to convert user queries into Cypher queries and to transform knowledge base responses into natural language. A sanitiser module was developed to verify the accuracy of the Cypher queries generated through LLM, significantly enhancing the system's overall reliability. The primary evaluation proved that transitioning from a traditional document-based format to a question-and-answering format and then to a dialogue-based format significantly enhances farmers' ability to access the desired information effectively and promptly.