Harnessing Generative AI for precision livestock management: methane emission reduction strategies in cattle farming

Abstract

Livestock farming is a significant contributor to global methane emissions, primarily through enteric fermentation and manure management, with disproportionate impacts on agricultural sustainability in developing regions like rural Kenya. This study addresses the critical gap between available mitigation strategies and their practical adoption by developing an AI advisory system combining evidence-based research with machine learning. The system implements a Retrieval-Augmented Generation (RAG) architecture utilizing three specialized components; Hugging Face’s all-MiniLM-L6-v2 model for 384-dimensional semantic embeddings of agricultural literature, Meta’s Llama3.2:1B model (deployed via Ollama) with constrained generation parameters (temperature=0, top_k=20) for reproducible outputs, and Chroma DB’s cosine similarity indexing for context-aware retrieval. Document processing incorporates fuzzy deduplication (90% similarity threshold) and semantic chunking (1000-character units with 100- character overlap) to optimize knowledge representation. Performance evaluation on 20 representative queries demonstrated 85% citation accuracy through manual Digital Object Identifier (DOI) verification, with 75% of responses rated as technically sound in preliminary developer assessments. While current local deployment on consumer hardware (Intel i5, 16GB RAM) yields 20-30 second response times, the architecture supports horizontal scaling through cloud hosting or edge optimization for field deployment. This work contributes a modular framework for climate-smart livestock management, demonstrating how lightweight AI systems can bridge research and practice. Future integration pathways include Tier 3 emission methodologies and offline capabilities for connectivity-limited regions, advancing both climate targets and agricultural equity. Keywords: Greenhouse Gases Emissions, Generative AI, Enteric Emmisions, Machine Learning.