Jona Rephisti

Jona Rephisti

Auto-generated profile, based on publicly available information

Jona Rephisti

Digital Green, Washington, USA

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
26 January 2026 in Advancements in Agricultural Development
Verified
Journal Article
Open access
VSVineet Singh
Vineet Singh
SSSanyam Singh
Sanyam Singh
LPLakshmi Pedapudi
Lakshmi Pedapudi
WPWaseem Pasha
Waseem Pasha
EKEkta Kumari
Ekta Kumari
EFEric Firnhaber
Eric Firnhaber
Recent generative AI offers personalized, high-quality advice to smallholder farmers in resource-limited settings. Yet, most large language models (LLMs) lack training data for diverse agroecologies, often yielding generic, inaccurate, or locally misaligned advice. Digital Green adapted Reinforcement Learning from Human Feedback (RLHF) to agricultural advisory to deliver highly localized, relevant, information. This refined tool, called Farmer.Chat, is an AI assistant supporting over 670,000 farmers in India, Kenya, Ethiopia, and Nigeria with text, image, and voice-based content. This paper details Digital Green's RLHF approach: a web-based annotation tool, multi-phase implementation, and quality assurance. Over 25,000 expert-reviewed Q&A pairs yielded significant improvements in response quality, tone, context, and cultural fit, especially for region-specific agricultural queries. The work outlines key lessons, cost/equity, and replication guidance. It calls for researchers, governments, and NGOs to pool validated Q&A data, strengthening global AI systems. Future work explores multimodal RLHF (image, voice, video), aiming to foster a global, inclusive, evidence-based ecosystem for AI agricultural advice.