AgriSight: Edge-to-Cloud Crop Diagnosis

# AgriSight: Offline-First AI for Smart Farming AgriSight is an AI-powered assistant that helps farmers quickly diagnose crop diseases, pest infestations, soil issues, and livestock health problems. Farmers simply capture a photo and optionally add a voice or text description. The solution uses a **two-tier AI architecture**. **Tier 1** is a lightweight **Gemma-class model** optimized for **AMD ROCm/Instinct MI300X**, designed for fast, low-cost diagnosis of common agricultural problems with a confidence score. If the confidence is low or the case is complex, it is automatically escalated to **Google Gemini** through Google AI Studio. Gemini combines the image with contextual information such as weather, location, crop stage, and user input to generate a detailed diagnosis, treatment recommendations, urgency level, and preventive guidance. This ensures powerful multimodal reasoning is used only when necessary. The key innovation is intelligent AI routing. Instead of sending every request to an expensive cloud model, AgriSight handles routine cases locally and reserves Gemini for difficult ones. As more verified diagnoses are collected, they are used to improve the lightweight model, enabling more cases to be resolved locally over time. This reduces cloud costs while improving accuracy and scalability. The application is designed for farmers with a simple interface, multilingual support, voice assistance, text-to-speech, and clear offline/online indicators. The hackathon prototype is built using Google AI Studio, while the production version will deploy the lightweight Gemma model on AMD hardware for true offline inference. By combining efficient edge AI with advanced multimodal reasoning, AgriSight delivers affordable, accessible, and scalable agricultural intelligence for farming communities worldwide.