Pezego-HITL: A policy-grounded large language model architecture for agricultural extension in Ghana

2026-07-15

AI Summary

This study addresses the multidimensional constraints—pertaining to safety, practicality, response latency, and expert burden—that hinder crop protection decision-making among smallholder farmers in Ghana. To tackle these challenges under stringent policy requirements, the authors propose Pezego-HITL, a policy-constrained large language model architecture that integrates structured retrieval-augmented generation with a verification memory routing mechanism, thereby ensuring regulatory compliance while enhancing recommendation utility and inference efficiency. The work further introduces P-EVAL, a novel evaluation framework that jointly models policy alignment, agronomic utility, response latency, and expert workload, enabling explicit trade-off analysis. Experimental results on 1,240 simulated queries demonstrate a policy alignment rate (PAR) of 0.94 and an agronomic utility rate (AUR) of 0.95, with P95 latency reduced by 55% to 12.9 seconds. Field evaluations involving 30 extension agents and 36 farmers confirm the system’s real-world usability.

Abstract

Large language models are increasingly deployed in agricultural decision-support settings, yet high-stakes crop protection in smallholder agriculture requires more than output-quality benchmarks. Over a two-year design and evaluation programme, we formalise policy-constrained large language model assessment as an adaptive compute allocation problem that jointly captures safety compliance, helpfulness, operational latency, and expert supervision workload. We introduce P-EVAL (Policy-grounded Expert-calibrated VALidation protocol), a unified evaluation framework for policy-grounded decision support, evaluating the architecture on a simulated field query database consisting of 1,240 cases. The protocol is instantiated on the Pezego advisory architecture (Pezego-HITL) and evaluated in Ghana. Following offline judge calibration against gold-standard human expert decisions ($κ= 0.77$), we evaluate the architectural performance under simulated query workloads. Under P-EVAL, our memory-routed architecture improves the Policy Alignment Rate (PAR) to 0.94 and the Agronomic Utility Rate (AUR) to 0.95, while reducing P95 latency by 55% (from 28.6s to 12.9s) through a 59.6% cache reuse ratio. We also demonstrate generalisability using the open-source \texttt{Qwen3.5-9B-DeepSeek-V4-Flash} model, achieving a PAR of 0.86 and a 54.5% latency reduction (to 10.2s). To evaluate practical utility and socio-technical integration, we administer detailed questionnaires to Ghanaian Extension Services Officers ($N=30$) and smallholder farmers ($N=36$). Taken together, this work demonstrates how policy-grounded structured retrieval-augmented generation with validated-memory routing makes safety-utility-latency trade-offs explicit, offering a scalable template for trustworthy AI-driven extension in smallholder farming systems.

Problem

• policy-grounded AI
• agricultural extension
• smallholder farming
• safety-compliance
• human-in-the-loop

Innovation

• policy-grounded LLM
• memory-routed RAG
• adaptive compute allocation
• P-EVAL framework
• agricultural extension