AgriSaathi: A Comprehensive Survey on Large Language Models and Multimodal AI for Agricultural Advisory Systems

Abstract

Abstract—The agricultural sector faces unprecedented challenges in meeting global food security demands while addressing climate change, resource constraints, and knowledge dissemination gaps. AgriSaathi represents a paradigm shift in agricultural advisory systems by leveraging Large Language Models (LLMs), multimodal AI, and Retrieval-Augmented Generation (RAG) to provide personalized, context-aware guidance to farmers. This survey paper comprehensively examines the evolution of agricultural advisory systems, from traditional extension services to modern AI-driven platforms. We analyze the role of LLMs in transforming agricultural knowledge dissemination, explore multimodal integration of vision, speech, and language processing for crop disease detection and farmer interaction, and investigate RAG frameworks for enhancing domain-specific agricultural knowledge retrieval. The AgriSaathi framework integrates a multilingual LLM-RAG core and vision-based crop diagnosis achieving up to 93–97 accuracy in agricultural query response and disease detection, based on results from benchmark models such as AgroLLM and Krishi Saathi.The paper critically evaluates existing systems, identifies technical and socio-economic challenges including low-resource language support and digital literacy barriers, and proposes future research directions. AgriSaathi synthesizes these technologies into a unified platform designed specifically for Indian farmers, addressing language diversity, regional crop variations, and real-time environmental data integration. Our findings indicate that LLM-based multimodal systems can significantly improve agricultural productivity and farmer decision-making when properly adapted to local contexts.