Author: WUR, FAO, Digital Green, Kissan AI and Ubuntoo

Publish Date: 14 April 2025

The Digital Agri Hub’s webinar on “ Unlocking Generative AI for Agriculture in LMICs ” brought together perspectives from the Food and Agriculture Organization of the United Nations(FAO), Digital Green, Ubuntoo, Kissan AI, and the Digital Agri Hub to explore how GenAI can transform food systems in low- and middle-income countries. From AI-powered advisory services and multilingual chatbots to context-aware LLMs, the session highlighted real-world tools, ethical challenges, and scaling strategies. A strong call emerged for cross-sector collaboration, curated data, and inclusive design.

Transforming Agriculture in LMICs with GenAI: What We Learned

On April 3, 2025, the Digital Agri Hub hosted a webinar exploring the potential of generative AI (GenAI) to tackle pressing challenges in agriculture across low- and middle-income countries (LMICs). From deploying Large Language Models (LLMs) to building chatbots, to curating local knowledge for food safety, the 90-minute session offered insights from developers, researchers, and practitioners at the forefront of digital innovation.

Sander Janssen from Wageningen University and Research (WUR) opened the session by framing the critical question: What does generative AI mean for agriculture in LMICs, and how do we use it responsibly? With the arrival of ChatGPT 2 years ago, Large Language Models (LLMs) have raised a lot of attention, and a lot of different implementations have been developed (e.g. Copilot, LLAMA, GROC, DeepSeek). Digital Solution providers are using these LLMs to build new solutions for small-scale producers, but other innovations are also available in providing policy advice by summarising a lot of information quickly. With the potential of LLMs and GenAI becoming widely available, it is timely to investigate their potential for agricultural applications.

The following applications of LLMs were presented in the Webinar:

Digital Green: Empowering Farmers Through Farmer.Chat

Jona Repishti presented Farmer.Chat , an AI-enabled multilingual chatbot currently used by 165,000 farmers across India, Ethiopia, Nigeria, and Kenya. Built for low-literacy users, it supports voice, text, and image inputs, delivering location- and season-specific advice. Early results are promising. Over 30% of users in Kenya report practice adoption, and many cite improved confidence and time savings.

Digital Green’s vision? Reducing advisory costs up to $0.35 per user (compared to $35/farmer by extension workers) while enabling peer learning, personalised nudges, and integrating marketplace access—all grounded in responsible AI development through human-in-the-loop evaluation and public datasets on Hugging Face.

Ubuntoo: Enhancing Food Safety through Collective Intelligence

Peter Schelstraete showcased the Food Safety for Africa project, combining AI-driven knowledge mining with curated expert input. By building a multilingual, contextual knowledge base—including tacit insights in local languages, Ubuntoo aims to close the digital divide between large and small farms.

Using retrieval-augmented generation (RAG) over a highly curated dataset, Ubuntoo’s platform eliminates hallucinations and offers reliable insights for researchers and, eventually, policymakers and informal market actors. They're also exploring AI agents to access multiple data sources, pushing the frontier of intelligent food safety systems.

Kissan AI: Scalable Voice-Based Copilots and Domain-Specific LLMs

Pratik Desai took us into the startup world, where Kissan AI has developed multilingual, voice-first copilot systems designed for farmers with limited digital literacy. Their innovations include:

• Vertical-specific LLMs trained on over 1.6 million data points.
• A market chatbot that provides dynamic insights on local crop prices.
• A knowledge graph-based system that reduces hallucinations.
• An enterprise copilot model used by Fortune 500 AgriTech firms.

Kissan AI is also building climate-resilient agriculture models and contributing to the open-source Agri Benchmark consortium alongside Bayer and UIUC.

FAO: Scaling AI Through Ethics and Digital Public Goods

Henry van Burgsteden from the Office of Innovation in FAO emphasised the organisation's commitment to ethical AI, data equity, and digital public infrastructure. Through initiatives like AgriLLM and partnerships with Digital Green, they aim to democratise GenAI for farmers and extension agents while advocating for transparency and inclusion.

FAO’s road map includes:

• Supporting digital advisory services and early warning systems.
• Empowering farmers through public datasets.
• Investing in dialogues, challenges and roadmaps via events like Reboot the Earth and the Digital Agriculture and AI Innovation Dialogue.

Digital Agri Hub: From Static Reports to Live Intelligence

Inder Kumar demonstrated the Hub’s latest innovation: Agri AI Chat, a chatbot that synthesises real-time data and 50+ curated knowledge products. Using RAG, LLMs, and prompt engineering, the assistant provides up-to-date insights to D4Ag stakeholders.

What sets it apart? A human-centred approach, integrated APIs from the Digital Agri Hub dashboard, and rigorous prompt testing to minimise hallucination. It’s a public good designed for transparency, accessibility, and global reach for market and business intelligence on the state of the digital agriculture sector.

Key Takeaways

• GenAI is already powering real tools: Generative AI is being actively used in LMIC agriculture. Tools like chatbots, voice assistants, and context-specific advisory platforms support farmers and agribusinesses. These technologies enhance access to knowledge, reduce advisory costs, and adapt to local needs, demonstrating real-world potential for scalable, data-driven agricultural innovation.
• Bias and hallucinations are not just bugs: AI hallucinations reveal underlying data or model gaps. Rather than ignoring them, organisations use expert reviews, domain-specific training, and red-teaming (intentional security testing of models) to improve outputs. These methods ensure safer, more accurate responses, especially where incorrect advice could harm crops, livestock, or farmer livelihoods.
• Context and inclusion matter: AI tools must reflect LMIC realities, including local languages, cultural norms, and digital access limitations. Inclusive design, user trust, and incorporation of indigenous knowledge are essential for effectiveness. Without this, GenAI risks excluding the very populations it aims to support in agricultural transformation.
• We need public infrastructure and shared benchmarks: Scaling GenAI in agriculture requires open, interoperable systems. Initiatives like AgriLLM and AgriBenchmark promote shared data standards and ethical AI use. Public infrastructure and digital public goods enable inclusive innovation, reduce duplication, and ensure LMIC stakeholders can develop, adapt, and benefit from AI equitably.

A call to action: Join the Next Webinar!

Webinar: Exploring the Potential of GenAI in Regenerative Agriculture Design
Date: April 22, 2025

Register via this link HERE.

In this second webinar in the series, we will explore how GenAI can support regenerative agriculture by addressing design complexity, data quality, indigenous knowledge, and collaboration frameworks. Whether you’re an AI developer, AgriTech innovator, or sustainability expert, this is your chance to co-create the future of inclusive, ethical, and regenerative AI in agriculture.