AgriTalk: A Trustworthy and Risk-Aware Bilingual AI Ecosystem for Multi-Crop Farm Management

Abstract

Systemic inefficiencies exist throughout the Indian agricultural industry, especially within the multi-crop regions of Maharashtra, due largely to their reliance on analog (manual, notebook-based), record keeping systems. These analog systems result in poor data accuracy, slow response time for decision- making, and no real-time visibility into farm profitability. This research study describes AgriTalk as a bilingual (English and Marathi) voice-assisted farm management ecosystem that uses large language models and retrievable augmented generation technology to professionalize smallholder farming operations. Furthermore, AgriTalk makes it easier for smallholder farmers to transition to digital technology with the provision of automatic labor tracking and granularity of plot management, along with providing complete traceability of value- added products such as raisins. Additionally, AgriTalk implements India’s 7 Sutras of Trust and the Digital Personal Data Protection Act (2023) to illustrate how AI-based, risk-aware architectures can offset the harmful effects of linguistic bias and model hallucination. Overall, preliminary research indicates that there are no longer inaccuracies in automated wage calculations as relate to piecework records; instead, there is now complete accuracy in computing automated wages and an elimination of administrative costs from maintaining manual records in comparison to manual record-keeping systems.