Cultivate: A Configurable AI Agent for Scaling Agricultural Extension Services in Ghana
DOI:
https://doi.org/10.65890/dmp-lncse.ICICCS26.203Keywords:
agricultural extension services, configurable AI agents, retrieval-augmented generation, Agent-as-a-Service, urban farming, human-in-the-loop, Ghana, knowledge transferAbstract
Ghana's agricultural extension system operates at a farmer-to-extension officer ratio of 1:1,500 three times the FAO-recommended standard of 1:500. Agronomists cannot serve the farmers who need them, and no existing system lets them train and deploy their own configurable AI agents. This paper presents Cultivate, an AgroAgent-as-a-Service (AaaS) platform that enables agronomists to upload domain knowledge, configure agent behaviour, and deploy AI agents that answer farmer queries at any time, unlike Farmer. Chat and FarmerAI, which use centrally curated knowledge bases, Cultivate gives each agronomist direct control over their agent's knowledge, communication style, and escalation threshold. The platform is built on Next.js, Mastra AI, Claude Sonnet 4.5, and Supabase with pgvector for retrieval-augmented generation (RAG). A pilot with Farmitecture a Ghana-based urban farming startup with 70+ customers and two agronomists validated the system with one agronomist and one farmer. The agronomist set up a fully configured agent in under ten minutes without technical guidance and estimated that the platform could handle 50-60% of his query load. The farmer said responses felt specific and grounded. The pilot demonstrated that decentralised, agronomist-trained AI agents are technically feasible and well-received by their intended users. Limitations include a small pilot sample, heuristic confidence scoring, and no WhatsApp integration yet.
Downloads
Published
Conference Proceedings Volume
Section
License
Copyright (c) 2026 DMPedia Lecture Notes in Computer Science & Engineering

This work is licensed under a Creative Commons Attribution 4.0 International License.