AgriLLM writeshops: Co-creating the future of agricultural intelligence

AgriLLM is a pioneering initiative that harnesses the power of large language models (LLMs) to accelerate digital transformation in agriculture. By integrating artificial intelligence with agricultural expertise. AgriLLM aims to empower farmers, extension agents, researchers, and policymakers with context-aware, multilingual, and user-friendly tools.

As part of this journey, DTA and IRRI organized two writeshops: a virtual session on 25 April and an in-person session on 30 June. These events provided dynamic platforms for collaborative content creation, focusing on building the question-and-answer (Q&A) pairs essential for fine-tuning AgriLLM.

Why a writeshop?

Unlike conventional workshops, the writeshops emphasized co-creation. Participants, including scientists, extension experts, and agricultural practitioners, worked together to develop context-specific Q&A datasets. These datasets form the backbone of AgriLLM, ensuring that the model reflects real-world challenges, terminologies, and decision-making contexts in agriculture.

The objectives the writeshops were clear:

• Generate high-quality Q&A content tailored to farmers, extension agents, and researchers.
• Advance domain-specific language for agricultural applications.
• Strengthen collaborative networks while fostering shared ownership of AgriLLM's development.

Key outcomes

• A contextualized Q&A dataset aligned with the needs and language of agricultural stakeholders.
• Identification of key agricultural challenges and mapping of stakeholders best positioned to address them.
• Improved understanding of AgriLLM's use cases and its potential to support real-time, evidence-based agricultural decision-making.
• Strengthened networks for ongoing collaboration and knowledge sharing.

Challenges and reflections

• Addressing hallucination risks in AI outputs through expert validation.
• Balancing scientific accuracy with user-friendly language for diverse audiences.
• Ensuring localization across geographies and languages.

Participants reflected positively on both modalities. While the virtual format enabled inclusivity and efficiency, the in-person session fostered richer dialogue, real-time feedback, and a deeper sense of trust.

The road ahead

The writeshops mark only the beginning of AgriLLM's journey. In the future, the initiative will:

• Strengthen data quality and inclusivity in multiple languages.
• Build strategic partnerships with research centers, universities, and innovation networks.
• Embed responsible AI principles such as transparency, bias mitigation, and equitable data stewardship.

By bringing diverse voices together, AgriLLM is not just building a digital tool, it is shaping an inclusive, participatory, and future-ready agricultural intelligence system.