Agri-Build

Inspiration

India feeds 1.4 billion people, yet the small-scale rural farmers who grow this food are often the most exploited. In regions like Fatehpur, up to 70% of a crop's value is lost to middlemen due to information asymmetry. Farmers sell their yields blindly, unaware of real-time mandi prices. Compounding this, crop diseases frequently wipe out entire fields because farmers lack access to agricultural experts.

I wanted to bridge this gap by putting AI directly into the hands of the farmers, enabling them to protect their crops, predict market trends, and sell directly for a fair price.

What it does

AgriBridge is an all-in-one AI platform designed specifically for small-scale Indian farmers:

• AI Crop Disease Detection: Farmers take a quick photo of a crop leaf and receive a 95.2% accurate diagnosis along with specific chemical/organic treatment recommendations in 5 seconds.
• Price Intelligence: Real-time prices from 500+ mandis combined with a 7-day price forecasting graph, helping farmers know exactly when and where to sell.
• Direct Marketplace: An open, transparent crop listing system where farmers connect directly with verified buyers, eliminating middlemen entirely.
• Secure Payments: Integrated with escrow-like payments via Razorpay to protect transaction values until delivery is verified.
• Progressive Web App (PWA): Installs like a native app and works on budget devices, even with spotty rural connectivity.

How I built it

I built AgriBridge as a solo developer in 36 hours:

• Frontend: Built with React.js and Tailwind CSS as a mobile-first Progressive Web App (PWA), deployed on GitHub Pages.
• Backend: Powered by Node.js and Express.js, using Firebase Auth for simple mobile OTP signups, and MongoDB Atlas for database management.
• ML Engine: A Python Flask service running two custom models:
  1. Disease Detection: A lightweight MobileNetV2 image classifier trained using transfer learning on the PlantVillage dataset (54k+ images covering 38 disease classes).
  2. Price Forecasting: An LSTM (Long Short-Term Memory) recurrent neural network trained on 5 years of historical mandi data from the government's Agmarknet database.
• APIs: Integrations with Agmarknet (live pricing), OpenWeatherMap (hyperlocal weather), and Razorpay (secure escrow payments).

Challenges I ran into

Building a complete, production-ready full-stack application and training two distinct machine learning models solo in 36 hours was intense. My primary hurdles were:

1. Lightweight ML: Ensuring the disease classifier was lightweight enough (using MobileNetV2) to eventually execute efficiently on low-cost, low-bandwidth smartphones.
2. API Data Wrangling: Dealing with the nested structures of the government's Agmarknet live price API feeds and transforming historical data into suitable time-series tensors for the LSTM network.
3. Offline Capability: Crafting service workers and local database sync logic to ensure farmers can draft crop listings and cache pricing data even when network connection drops out in fields.

Accomplishments that I'm proud of

• Designing and implementing a clean, fully functional PWA that runs seamlessly on both desktop and mobile devices.
• Achieving a 95.2% accuracy rate (0.94 F1-score) on the crop leaf disease classifier.
• Building a complete end-to-end user flow: from capturing a leaf photo to securing direct buyer matching and escrow payment, all single-handedly.

What I learned

• How to prepare, clean, and structure time-series agricultural data to train LSTM neural networks.
• How to build responsive, lightweight PWAs that focus heavily on offline availability and low-data consumption.
• The business and logistical realities that small-scale rural farmers face daily, which helped shift my focus toward simplistic UI designs.

What's next for AgriBridge

• Voice Integration: Implementing a multilingual voice interface (Hindi, Bhojpuri, etc.) so farmers who cannot read or write can easily navigate the app using natural speech.
• USSD / SMS Fallback: Adding a simple text-message-based fallback system for older feature phones.
• Satellite Scanning: Integrating remote sensing crop health diagnostics via satellite imagery.

Built With

• agmarknet
• api
• express.js
• firebase
• flask
• github
• javascript
• keras
• lstm
• machine-learning
• mobilenetv2
• mongodb
• node.js
• pwa
• python
• razorpay
• react
• render
• tailwind-css
• tensorflow