krishi sakhi

some screenshot of our working prototype

Inspiration

Deep gaps in digital awareness, erratic crop results, and restricted access to real-time analytics continue to plague India's farming community. Our goal was to provide small and mid-sized farmers with an AI-powered friend, or sakhi, who would assist them in making informed decisions based on facts rather than intuition.

What it does

The following features are offered by Krishi Sakhi, an intelligent agri-assistance platform: 🌾 Crop Health Detection: Farmers can upload or take pictures of their crops, and our AI model—powered by Google Gemini Vision + Teachable Machine—identifies illnesses and suggests treatments. ☁️ Smart Weather & Soil Analytics: Linked with IoT soil sensors and OpenWeather API to provide real-time data and predictions powered by AI. 💬 Voice-Based Interaction: Google Speech-to-Text and the Gemini multimodal API enable farmers to communicate with the app in their native tongues. 💰 Market Price Tracker: Firebase Cloud Functions + Gemini-generated insights combined with real-time Mandi data to anticipate price patterns. 🧠 Personalized Advisory: The backend retrieves training materials and advisories tailored to a given location from our knowledge base and AI models using GenKit workflows. 🔒 Smart Offline Access & Authentication: Secure local caching and Supabase auth for remote locations with poor connectivity.

How it was constructed Frontend: Tailwind UI for the web version, Flutter (mobile first). Backend: Node.js using GenKit pipelines to coordinate IoT feeds, weather data, and the Gemini API. Supabase (PostgreSQL) is the database used to store analytics, crop logs, and user data. AI Layer: TensorFlow Lite for on-device disease categorization, Teachable Machine, and Google Gemini API.

The difficulties we encountered

Without hallucinations, tune Gemini for agriculture advise specific to a given locale. coordinating IoT and AI data while managing low-bandwidth situations. establishing precise crop-disease databases for crops grown in India. putting in place voice commands in multiple languages that functioned flawlessly offline.

Achievements of which we are proud GenKit and Gemini multimodal were successfully integrated with real-time text and image processing. created a voice-first user interface that is fully functional and usable by non-tech farmers. Under patchy internet, a sub-2-second response time was attained for live crop diagnosis.

What we discovered

the potential of integrating edge computing, IoT, and AI for practical rural effect. Quality of the dataset is more important than model size. practical knowledge of Supabase edge functions, Gemini API, and GenKit pipelines. How usefulness in field conditions is affected by minor latency optimizations.

Future Plans for Krishi Sakhi For convenience, integrating a WhatsApp chatbot.

collaborating with nearby cooperatives to bring early adopters on board.

establishing a Krishi Insights Dashboard to monitor crop health and climatic impact in different regions.