Empowering Smallholder Farmers with AgriLLM: A Low-Latency, High-Accuracy AI Assistant for Context-Relevant Farming Advice

Smallholder farmers lack access to localized agricultural advice, while generic AI tools often provide unsafe or irrelevant recommendations. This paper introduces AgriLLM, an AI assistant that delivers accurate, context-aware farming guidance by combining large language models with region-specific agricultural databases. Tested on 500 queries from Indian farmers, AgriLLM achieves 89.6% accuracy (vs. 64.2% for standard GPT-4 [2]) and reduces incorrect advice by 81%. The system runs on budget smartphones with <2 sec response times, making it practical for rural areas. Field trials with 112 farmers showed 92% satisfaction, with particular praise for its crop-specific pesticide warnings and irrigation advice. AgriLLM demonstrates how AI can empower farmers when grounded in local knowledge, offering a scalable solution to global food security challenges. Future work will expand support for voice inputs and rare crops.