Skip to main content Skip to navigation

AgroLLM: Connecting Farmers and Agricultural Practices through Large Language Models for Enhanced Knowledge Transfer and Practical Application

Project Overview

The document examines the implementation of AgroLLM, a specialized large language model (LLM) aimed at enhancing education and knowledge sharing in the agricultural sector. By employing a Retrieval Augmented Generation (RAG) framework, AgroLLM delivers precise and contextually relevant responses to agricultural inquiries, leveraging a vast database of agricultural resources. The evaluation of various LLMs, including ChatGPT-4o mini, Gemini 1.5 Flash, and Mistral-7B-Instruct, indicated that ChatGPT-4o mini outperformed the others in accuracy and response speed. The project's primary goal is to bridge the divide between theoretical agricultural knowledge and its practical application in the field. Additionally, the document addresses challenges faced in this endeavor, particularly concerning data quality and the need for effective domain adaptation. Overall, the findings underscore the potential of generative AI in transforming agricultural education by providing tailored, real-time support to learners and practitioners.

Key Applications

AgroLLM chatbot

Context: Educational tool for farmers and agricultural professionals seeking knowledge and practical application in agriculture.

Implementation: Developed as a chatbot using a large language model framework, integrating data collection, embedding generation, and retrieval-augmented generation.

Outcomes: Enhanced knowledge sharing, improved decision-making, and practical applicability in agriculture, with a reported accuracy of 93% for the best-performing model.

Challenges: Challenges include the need for domain adaptation, data quality issues, and skepticism from users regarding automated systems.

Implementation Barriers

Technical Barrier

LLMs may not fully capture specialized agricultural vocabulary and concepts without fine-tuning on agriculture-specific datasets.

Proposed Solutions: Fine-tune LLMs on agriculture-specific datasets to improve contextual understanding.

Data Barrier

Agricultural data is diverse, coming from various sources and formats, which makes quality control difficult.

Proposed Solutions: Expand the dataset to include a broader range of sources, including field reports and real-time sensor data.

User Trust Barrier

Farmers and agricultural experts may be skeptical of the outputs from LLMs due to potential inaccuracies.

Proposed Solutions: Incorporate feedback mechanisms to refine response quality and enhance user trust.

Project Team

Dinesh Jackson Samuel

Researcher

Inna Skarga-Bandurova

Researcher

David Sikolia

Researcher

Muhammad Awais

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Dinesh Jackson Samuel, Inna Skarga-Bandurova, David Sikolia, Muhammad Awais

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies