Empowering Indian Farmers with Multilingual AI: A Voice-Enabled ChatBot Using Dhenu2
Biswayan Mehra, J Anitha · 2025
The linguistic and technological diversity of India poses unique challenges to equitable availability of agricultural information. A multilingual, voice-enabled Farming ChatBot is proposed in this research, to bridge the gap between farmers and domain-specific Artificial Intelligence systems. At the core of the system is Dhenu2, an 8-billion-parameter large language model developed by KissanAI, specifically trained on Indian agricultural data and farmer-centric use cases. While Dhenu2 natively supports bilingual interaction in English and Hindi, this work extends its capabilities to eleven Indian languages through a modular pipeline leveraging SarvamAI's speech-to-text and text-to-speech APIs. The system architecture comprises of three core modules: (i) multilingual voice/text input translation, (ii) Dhenu2-based domain reasoning and response generation, and (iii) output rendering in the user's native language via both text and audio. The entire pipeline is deployed within a lightweight React Native mobile application optimized for low-latency, ondevice usage. Evaluation of the system is conducted using a curated CSAT benchmark derived from the Government of India's Kisan Call Center dataset. The ChatBot records an average CSAT rating of 4.1 out of 5, with an overall agreement of 81% between the model and expert responses. In addition to demonstrating Dhenu2's robustness over other agricultural models such as KissanGPT and AgriBERT, the proposed system showcases a scalable architecture suitable for real-world deployment. Future enhancements include offline support and integration with Dhenu-Vision for plant disease diagnosis from image input.