Algoverse

Mydrak AI – Intelligent, Cost-Aware Crop Health Assistance Smallholder farmers often face crop losses because timely agricultural expertise is difficult to access, expensive, or unavailable in their local language. Mydrak AI aims to bridge this gap through an AI-powered agricultural assistant designed to make preliminary crop-health guidance more accessible. Built around a Telegram-based interaction model, Mydrak AI is designed to let farmers share photographs of affected plants and receive structured, understandable agricultural guidance without installing a separate application. What makes Mydrak AI different is its cost-aware, multi-model AI architecture. Instead of sending every request to an expensive model, it uses a tiered reasoning approach involving Muse Glimmer 30B for cost-efficient preliminary analysis and Muse Spark 1.3 for more demanding reasoning. Its architecture also incorporates Meta's SAM 3.1 for planned plant-region segmentation and Muse Voice Transcribe for future voice-based interaction. The system includes image intake and preprocessing, diagnostic decision logic, confidence evaluation, and a structured agricultural knowledge base. It is designed to distinguish between sufficiently supported results and uncertain cases, requesting better information or recommending human expert review rather than presenting guesses as confirmed diagnoses. Another important component is its source-grounded agricultural knowledge base, which prioritizes institutional agricultural publications and region-specific information. The initial focus includes potato, soybean, and tomato-related resources, with particular attention to agricultural guidance relevant to Madhya Pradesh. Privacy and responsible AI are central to the architecture. Mydrak AI incorporates consent management, data-policy enforcement, restricted external processing, hashed identifiers, and controlled response generation. It also includes spending safeguards, bounded model calls, offline workflows, mock providers, cached replay, and automated tests to support reproducible development while minimizing unnecessary API usage. The project is being developed as a practical agricultural assistance system, not merely another general-purpose chatbot. Its objective is to combine advanced AI reasoning with affordability, agricultural knowledge, transparency, and responsible decision-making. Although live model integrations, end-to-end field validation, and expert-reviewed disease recommendations remain under development, the project establishes a modular foundation for bringing intelligent agricultural assistance closer to the farmers who need it. Our vision: Make agricultural intelligence more accessible, AI assistance more affordable, and every recommendation more responsible.