The agricultural landscape is undergoing a profound digital transformation, and at the heart of this revolution lies an artificial intelligence platform that is fundamentally reshaping how millions of smallholder farmers access critical knowledge. FarmerChat V2, developed by the nonprofit organisation Digital Green, represents a ground-up rebuild of an already successful AI-powered advisory assistant that has now reached over one million users across India, Kenya, Ethiopia, Nigeria, and Brazil. This upgraded version is not merely an incremental improvement but a comprehensive reimagining of how technology can bridge the persistent gap between agricultural expertise and the farmers who need it most.
Smallholder farmers around the world face a daunting array of challenges that make their livelihoods inherently volatile. Among the most significant obstacles is the difficulty of accessing timely, contextually relevant advice at the precise moment when critical decisions must be made. The knowledge exists decades of agricultural research, local agronomic wisdom, and climate-smart practices are well documented but reaching that knowledge when a farmer stands in a field confronting a pest infestation or an unexpected weather pattern has historically been extraordinarily difficult. Traditional agricultural extension services, while valuable, have struggled to scale effectively, with in-person advisory costs reaching approximately ₹3,300 per farmer annually, a figure that makes comprehensive coverage economically unfeasible for most developing nations.
FarmerChat V2 emerges as a direct response to this challenge, leveraging cutting-edge artificial intelligence to deliver personalised, hyperlocal farming guidance at a fraction of the cost approximately ₹33 per farmer per year. This represents a hundredfold reduction in the cost of agricultural extension, fundamentally altering the economics of knowledge dissemination and opening the door to reaching millions more farmers than conventional systems could ever accommodate.
The Evolution from V1 to V2
The journey to FarmerChat V2 began with the original platform’s launch in October 2024, which quickly demonstrated strong adoption and measurable impact. The first version saw 67 percent of active users applying the advice they received, according to a third-party evaluation conducted by IDinsight. However, the Digital Green team recognised that true transformation required more than simply providing answers it required understanding farmers’ evolving digital behaviours and adapting the platform accordingly.
Through millions of interactions, farmers themselves illuminated the path forward. A clear pattern emerged: users were moving away from typing their questions, choosing instead to speak to the assistant or photograph the problems they encountered in their fields. Notably, those who utilised the image-based diagnosis feature proved significantly more likely to act on the advice they received. This behavioural shift, combined with feedback from farmers who expressed a desire for advice tailored to their specific crop, season, and language, catalysed the development of a fundamentally different approach.
FarmerChat V2 represents what Digital Green describes as a “ground-up rebuild” of the assistant, one that is faster, more intuitive, and far better at understanding what a farmer is actually asking, whether they type, talk, or send a photo. The platform now asks clarifying questions before providing answers, mimicking the approach of a trusted local shopkeeper who takes time to understand a customer’s needs before offering solutions.
Key Features and Capabilities of FarmerChat V2
A. Multimodal Interaction
FarmerChat V2 supports three primary modes of interaction, each designed to accommodate the diverse communication preferences and literacy levels of smallholder farmers. First, text-based input allows farmers to type their questions, though this has become less common as voice and image features have gained prominence. Second, voice interaction enables farmers to speak naturally to the assistant, removing barriers related to literacy and typing proficiency. Third, and perhaps most significantly, photo-based diagnosis allows farmers to simply capture an image of a diseased plant, an unfamiliar insect, or any other visual concern and receive instant analysis and recommendations.
This multimodal approach is not merely a convenience it represents a fundamental recognition that many agricultural problems are easier to photograph than to articulate. A farmer encountering an unfamiliar pest may lack the vocabulary to describe it accurately, but a photograph captures all relevant details instantaneously. The V2 platform saves farmers’ preferences, learning whether they prefer speaking or uploading photos, and adapts its interface accordingly.
B. Personalisation and Adaptive Advisory
One of the most transformative features of FarmerChat V2 is its capacity for deep personalisation. The platform now maintains an evolving farmer profile shaped by each user’s specific crop, location, and seasonal requirements. This profile enables the system to surface suggested starter prompts tailored to what a farmer should be asking during their particular agricultural context, rather than simply waiting for questions to be formulated.
The advisory cards introduced in V2 represent a significant advancement in how information is presented. Rather than delivering a wall of text, the platform organises guidance into digestible, actionable cards that are easier to understand and implement. This is complemented by richer responses that include relevant images, making agricultural information more accessible and actionable precisely when it matters most.
C. Language and Accessibility
FarmerChat V2 is available in five languages, a deliberate design choice that recognises the linguistic diversity of the farming communities it serves. The platform works in any language a farmer speaks, with the underlying technology built to accommodate linguistic variety across different regions and cultures. This multilingual capability is particularly critical in countries like India, where farmers may speak languages that are not well-served by conventional digital tools.
The platform is also designed to function effectively in low-connectivity environments, acknowledging that many rural areas lack reliable internet access. This offline-capable design ensures that farmers can access critical information even when connectivity is intermittent or unavailable.
D. Behind the Technology: RAG, RLHF, and Expert Validation
The technical architecture underpinning FarmerChat V2 combines several advanced AI methodologies to ensure both accuracy and relevance. At its core, the platform employs Retrieval-Augmented Generation (RAG), which combines a curated knowledge base with a large language model to generate responses grounded in verified agricultural information.
The RAG system functions by first retrieving relevant documents from a regularly updated knowledge base composed of both structured and unstructured content, including materials vetted by government organisations and agricultural experts. The language model then generates natural language responses based on these retrieved documents, ensuring that outputs remain grounded in verified agronomy rather than potentially hallucinated information.
Complementing the RAG architecture is Reinforcement Learning from Human Feedback (RLHF), which refines the model’s responses based on evaluations from local agronomists and veterinarians. These experts regularly validate the responses farmers receive, feeding corrections back into the models so that answers stay accurate and locally relevant. This human-in-the-loop approach ensures that the AI remains a tool for democratising expert knowledge rather than replacing it.
The platform’s performance metrics reflect the effectiveness of this architecture. Evaluations have shown that FarmerChat achieves 71 percent context precision, ensuring that responses accurately reflect source materials. Furthermore, 80 percent of responses score above 0.7 for faithfulness, preventing hallucination and ensuring that advice remains grounded in verified sources. The system maintains 67 percent high relevance in matching farmer needs, with average response times of approximately 9.05 seconds, enabling real-time decision support during critical agricultural moments.
Measurable Impact and Adoption
The impact of FarmerChat V2 is perhaps best understood through the data generated since its rollout. Activation rates the proportion of users who download the app and go on to ask a genuine question climbed significantly across every major market measured, rising from around half of users to approximately two-thirds. This increase represents thousands of additional farmers engaging meaningfully with the platform rather than abandoning it after initial download.
More striking still is the depth of engagement. Farmers using V2 ask approximately one and a half times as many questions as they did on the previous version. Ethiopia has emerged as the clearest example of this trend, with farmers now asking around four questions per visit, up from approximately two and a half. This pattern suggests that farmers are not merely curious but are finding sufficient value in the responses to continue the conversation—a back-and-forth that indicates trust rather than passing interest.
The platform’s user base now exceeds one million farmers and frontline extension workers across India, with women making up approximately 45 percent of users. FarmerChat has answered over three million farming queries since its launch, addressing questions spanning crop planning, pest and disease management, livestock care, weather forecasts, and agricultural inputs.
Third-party evaluation conducted by 60 Decibels has documented the platform’s tangible impact on farming practices. According to this research, approximately 60 percent of active users act on the advice they receive, and 91 percent report greater confidence in their farming decisions. The study also found that 83 percent of users say accessing information is “much easier” through FarmerChat, and nearly three in four farmers rate the information as “very relevant” to their specific circumstances. The platform achieved an exceptional Net Promoter Score of 63, reflecting strong user satisfaction and trust.
In regions like Madhya Pradesh, India, farmers have successfully pivoted to organic pest control based on FarmerChat’s recommendations, leading to healthier soil and higher yields. This example illustrates how the platform translates artificial intelligence into practical, sustainable agricultural improvements.
The Economic Case for AI-Powered Extension
The cost-effectiveness of FarmerChat V2 deserves particular attention because it fundamentally changes the calculus of agricultural development. While traditional in-person advisory services cost approximately ₹3,300 per farmer annually, FarmerChat reduces this to approximately ₹33 per farmer per year. This hundredfold reduction is not achieved by compromising quality but by leveraging technology to scale expert knowledge in ways previously impossible.
The efficiency gains extend beyond direct costs. By fine-tuning smaller, highly efficient AI models rather than relying solely on massive frontier models, Digital Green and its partner OpenAI have created a sustainable economic model for the platform. This approach provides consistently high-quality responses while reducing infrastructure costs, ensuring the platform’s long-term viability.
“FarmerChat is making it easier and less expensive for agricultural extension agents to provide smallholder farmers in India with timely, accurate data about where to grow specific crops and how to protect them from drought and disease,” said Anna Makanju, Vice President of Global Affairs at OpenAI. This partnership between a nonprofit organisation and a leading AI company illustrates how cross-sector collaboration can address pressing global challenges.
The Road Ahead: From Advisory to Agency
The future roadmap for FarmerChat extends well beyond its current capabilities. Digital Green is preparing to integrate OpenAI’s Operator, an advanced AI agent that will transform FarmerChat from a tool that merely provides advice into a partner that takes action. This integration will enable the platform to complete tasks on behalf of farmers, bridging the gap between knowledge and implementation.
Imagine a farmer seeking to improve soil fertility and boost crop yields. Currently, FarmerChat assesses the farmer’s specific crop type, soil condition, and local weather patterns, then delivers tailored recommendations for specific fertiliser amounts along with detailed application instructions. With Operator, the platform will go further helping the farmer purchase the recommended fertilisers from local suppliers, displaying comparison pricing across a range of options, and simplifying the path from problem to solution.
“As we learn more about Operator during its research preview, the goal is for Farmer.Chat to become a true partner that bridges the gap between advice and action understanding individual farmers and connecting them to local solutions in real time,” said Rikin Gandhi, CEO of Digital Green.
Beyond Operator integration, Digital Green is developing what it calls the Agricultural AI Harness, an ecosystem foundation where trusted services for advisory, weather, markets, and government schemes come together, powered by reasoning and memory that understands each farmer’s context. This vision encompasses a comprehensive digital infrastructure for agricultural support, one that could fundamentally transform how farmers access information, markets, and services.
The organisation aims to add another 35 lakh farmers to the platform by 2028, with particular emphasis on strengthening women farmers’ participation. Expansion plans include rolling out FarmerChat access across five major regions in Ethiopia, deepening integration across channels, and refining a Hausa language version to reach Nigeria’s 38 million farmers.
Challenges and Considerations
While FarmerChat V2 represents remarkable progress, its development and deployment are not without challenges. The platform must continually adapt to local agricultural conditions, which vary dramatically across regions, crops, and seasons. Ensuring that AI-generated advice remains accurate and relevant across such diversity requires ongoing investment in knowledge base updates, expert validation, and user feedback mechanisms.
The platform also operates within a broader context of digital infrastructure limitations in many rural areas. While it is designed to function in low-connectivity environments, complete offline functionality remains constrained by the inherent limitations of AI systems that require cloud-based processing. Bridging this gap will require continued innovation in edge computing and offline-capable AI models.
Furthermore, the success of FarmerChat ultimately depends on farmers trusting and acting upon the advice provided. While the 60 Decibels evaluation indicates high levels of confidence and adoption, ongoing monitoring and evaluation are essential to ensure that the platform continues to deliver genuine value rather than simply generating engagement metrics.
Conclusion
FarmerChat V2 represents a milestone in the application of artificial intelligence to global agricultural development. By combining advanced AI technologies with deep local knowledge and continuous farmer feedback, Digital Green has created a platform that delivers trusted, hyperlocal, and actionable advice to millions of smallholder farmers at a cost that makes universal access economically feasible.
The platform’s success is measured not in downloads or engagement metrics alone but in the tangible improvements it enables: healthier soil, higher yields, greater confidence in decision-making, and ultimately, more resilient livelihoods for some of the world’s most vulnerable populations. As FarmerChat continues to evolve expanding to new regions, incorporating new capabilities, and deepening its integration into agricultural extension systems it offers a compelling model for how technology can be deployed in service of human flourishing.
The transformation of global farming that FarmerChat V2 represents is not merely technological but philosophical. It demonstrates that the most powerful applications of artificial intelligence may not be those that replace human expertise but those that scale it, making the wisdom of agricultural experts accessible to every farmer who needs it, whenever and wherever they need it. In this vision, technology becomes not a barrier but a bridge connecting farmers to the knowledge, tools, and services they need to thrive in an increasingly unpredictable world.
