Saajhedar: Multilingual Agricultural AI Scheme Assistant

Saajhedar, meaning “partner” in Hindi is a multilingual, voice-first conversational AI assistant designed to help rural Indian farmers access government agricultural schemes despite language and literacy barriers. The system integrates Automatic Speech Recognition (ASR), Natural Language Understanding (NLU), and agentic dialog management to enable farmers to ask questions, verify eligibility, and receive personalized scheme recommendations in Hindi, Marathi, and English. Built on a hybrid architecture combining Wav2Vec2, LangChain, and mT5 models, Saajhedar demonstrates the feasibility of a culturally grounded, low-literacy, AI-driven agricultural support system. Prototype evaluation achieved a Word Error Rate (WER) of 14.8% (Hindi), intent accuracy of 86.2%, and strong usability ratings, confirming the effectiveness of the adopted hybrid methodology. The integration of multilingual ASR and agentic dialog workflows directly enhanced system usability, establishing a foundation for future WhatsApp-based deployment and scalable rural digital inclusion.