Living Agricultural Intelligence: Dynamic AI Advisory for Rice Landscapes

Rice production systems are increasingly shaped by climate variability, fragmented extension capacity, and uneven access to high-quality agronomic intelligence. Traditional advisory models rely on static recommendations that cannot respond to real-time field variation or evolving management conditions. Farmers require advisory systems that learn continuously from landscape data and farmer interaction, converting passive datasets into active, context-aware intelligence. This initiative introduces a living advisory architecture designed to bridge scientific rigour, digital scale, and equitable farmer access.

The solution is a co-developed dynamic advisory system that converts passive data into active landscape intelligence. Its foundation is a dynamic data stack integrating Rice Crop Manager (RCM), Landscape Crop Assessment Survey (LCAS/LCAS+), and earth observation datasets. Historical agronomic and spatial datasets are combined with increasingly high-resolution geospatial data streams. Advanced AI/ML models analyse this combined dataset to generate spatial and context-specific inputs. Rather than producing generic summaries, the system infers how plot-level and management conditions shape the most appropriate recommendations. These inputs are fed into domain-specific AgriLLMs that ingest evolving data—including lean data submitted by farmers through chat interfaces—and continuously refine outputs. The system operates under the DynAg framework co-developed by IRRI to ensure scientific validity, social inclusion, and prioritisation of marginalised farmers and local languages.

Deployment follows a multi-modal delivery strategy. Intelligence is delivered through WhatsApp chatbots such as PaddyMitra and FarmerChatbot, and the government's Kisan Sarathi platform. A human-assisted AI model is used for quality assurance: AgriLLMs generate advisory content while human experts validate complex or high-stakes queries. A feedback loop captures lean data from farmer interactions and feeds it back into the intelligence stack, allowing the system to adapt continuously to changing field conditions.

AI/ML analytics-based advisories have already reached more than one million farmers through Jeevika and related extension platforms. LCAS+ further expands scale by mobilising agricultural students from state universities in partnership with ICAR and local institutions, creating a sustainable pipeline of ground-truth data collection. Measured impact shows a 36% reduction in nitrogen pollution through AI-targeted nutrient advice, and regional modelling indicates potential increases of 2.22 million tonnes in sustainable rice production through AI-guided irrigation strategies. Field trials report yield gains of up to 1 tonne per hectare from targeted productivity interventions. Ethics and governance follow the DynAg framework, treating farmer data as a sovereign asset used only to improve services. Multi-modal delivery ensures accessibility for low-literacy users across diverse geographies.
For additional context and detailed documentation of this use case, please refer to pages 27-28 in the attached Casebook.