Publication

Abstract : Farming in India continues to face difficulties due to a lack of digital platforms that support local languages and rural contexts, fragmented market information, and restricted access to expert guidance. Many existing agricultural tools remain task-specific, rely heavily on image data, or provide little interpretability, which restricts their practical use. To address these gaps, this paper presents AgriGPT, an integrated large language model framework that unifies four specialized modules into a single decision-support system. The translation module enables real-time multilingual interaction through speech and text, extending inclusivity beyond the text-only or English-focused systems found in prior work. The crop recommender leverages soil and climatic attributes with a Random Forest model to achieve 94.6% accuracy, surpassing conventional baselines such as Logistic Regression and SVM. The disease predictor, built on a fine-tuned DistilBERT, classifies plant diseases from symptom descriptions with 91.7% accuracy, offering a practical alternative where image-based classifiers are less effective. A fuzzy logic market advisor adds interpretability by balancing price, distance, and infrastructure to generate transparent suitability scores, a capability rarely addressed in earlier studies. Integrated through a lightweight Streamlit interface, AgriGPT demonstrates how domain-specific LLMs can combine accuracy, usability, and inclusivity to support real-world agricultural decision-making.