The Challenge
Farmers and extension workers navigate complex decisions every day, from crop health and weather variability to input use and market conditions.
While agricultural knowledge exists, access to timely, localized, and understandable advice remains uneven across languages, regions, and crop contexts, and public extension systems face capacity constraints in responding to the volume and diversity of farmer queries.
Our AI Solution
AgriVaani is a multilingual, multimodal conversational AI chatbot that delivers agricultural advisories to farmers and extension workers, powered by an agriculture domain-specific language model developed by Wadhwani AI.
Going forward, Users can ask questions via text or voice and receive contextual, localized, actionable responses in text and audio, grounded in curated agricultural knowledge sources rather than open-ended generation.
How it works
01
User submits a query using voice or text
A user submits a query using voice or text in their preferred language.
User submits a query using voice or text
A user submits a query using voice or text in their preferred language.
02
Query is translated
The query is translated into English.
Query is translated
The query is translated into English.
03
AgriVaani uses a retrieval-augmented
AgriVaani uses a retrieval-augmented generation (RAG) pipeline to retrieve relevant information from expert-curated datasets before generating a response.
AgriVaani uses a retrieval-augmented
AgriVaani uses a retrieval-augmented generation (RAG) pipeline to retrieve relevant information from expert-curated datasets before generating a response.
04
Response is translated
The response is translated back into the user's language and delivered in text and audio form.
Response is translated
The response is translated back into the user's language and delivered in text and audio form.
05
An agent-based architecture
An agent-based architecture - including query rephrasing and intent detection - routes queries appropriately and handles cases where a reliable answer cannot be provided.
An agent-based architecture
An agent-based architecture - including query rephrasing and intent detection - routes queries appropriately and handles cases where a reliable answer cannot be provided.
06
Safety and grounding
responses draw on Package of Practices issued by government bodies, historical Kisan Call Center Q&A data, and SME-curated benchmark datasets, with guardrails restricting unsafe advisories (including certain pesticide recommendations) and a human-in-the-loop process throughout development and evaluation.
Safety and grounding
responses draw on Package of Practices issued by government bodies, historical Kisan Call Center Q&A data, and SME-curated benchmark datasets, with guardrails restricting unsafe advisories (including certain pesticide recommendations) and a human-in-the-loop process throughout development and evaluation.
Where It’s Being Tested
Deployed in Madhya Pradesh and Uttar Pradesh. AgriVaani supports Hindi, Marathi, Punjabi, Gujarati, and English, with advisory coverage across ten crops.
Impact
AgriVaani supports everyday crop-care and input-use questions, seasonal and weather-related risk guidance, and frontline support for extension workers and helplines.
Our Other Solutions
Partners
Looking Ahead
Planned enhancements include expanding coverage to additional crops and languages, enabling multi-turn conversations, and incorporating image-based pest and disease advisory, alongside continued dataset development and rigorous evaluation for accuracy, relevance, and scale.