AgriBridge：以開源 AI 連接農業與智能

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

What if farmers could access agricultural intelligence as easily as sending a message?

Agriculture remains the backbone of millions of livelihoods, yet smallholder farmers often struggle with delayed crop disease identification, inaccessible expert guidance, language barriers, and unreliable internet connectivity.

To explore a solution to these challenges, I developed AgriBridge, an open-source, AI-powered agricultural intelligence platform designed for Indian farmers.

Rather than functioning as another generic AI chatbot, AgriBridge integrates computer vision, multilingual speech recognition, retrieval-augmented generation (RAG), and environmental data to deliver context-aware agricultural assistance.

🌱 What can AgriBridge do?

• AI Crop Assessment: Uses a DINOv2-based ONNX classification pipeline to analyze crop leaf images and provide preliminary crop-health assessments.
• Multilingual Advisory: Supports agricultural interactions through voice and text in English, Hindi, and Bengali.
• Field-Aware Intelligence: Combines agricultural knowledge with crop information, weather conditions, soil properties, and geographical context.
• Offline-Resilient Scanning: Uses IndexedDB to queue crop scans locally when internet connectivity is unavailable and processes them after reconnection.
• AI-Powered Knowledge Retrieval: Incorporates BGE-Small embeddings and RAG to retrieve relevant agricultural information.
• Sustainability Explorer: Provides illustrative carbon and biomass estimations to encourage awareness of environmentally responsible agricultural practices.

💰 Who would pay for it?

AgriBridge's potential customers include Farmer Producer Organizations (FPOs), agricultural cooperatives, farming enterprises, and agricultural advisory service providers.

A possible business model involves providing farmers with accessible basic tools while offering organizations premium services such as multi-farm management, extended analytics, and consolidated advisory dashboards.

The commercial model is currently conceptual and has not yet been validated with paying customers.

The goal is to make agricultural intelligence more accessible while developing a potentially sustainable business around practical farming assistance.

Demo

🚀 Live Application: Explore AgriBridge

AgriBridge provides a responsive, mobile-first interface through which users can explore agricultural intelligence features.

How it works

1. Get Started: Access AgriBridge and select a preferred language.
2. Configure Your Farm: Add crop, location, and available field information.
3. Interact: Upload a leaf image or ask an agricultural question using voice or text.
4. AI Processing: Backend services coordinate crop classification, speech transcription, knowledge retrieval, and environmental context.
5. Receive Guidance: Review the generated agricultural assessment, precautions, and contextual recommendations.

Offline capability: Crop images can be queued locally during connectivity interruptions. AI analysis requires connectivity to the backend.

Code

The project is publicly available on GitHub.

💻 GitHub Repository: github.com/8ernity/Agri-Bridge

The repository includes the application frontend, backend services, AI integrations, configuration files, and deployment resources.

AgriBridge is open to contributions, technical improvements, and collaboration from developers interested in agricultural technology.

How I Built It

AgriBridge follows a modular architecture that separates the user interface, backend orchestration, AI intelligence services, and agricultural data integrations.

⚛️ Frontend — React, Vite and Tailwind CSS

The user interface is developed with React and Vite, using Tailwind CSS for responsive styling.

A Progressive Web App approach provides a mobile-friendly experience with browser-based storage capabilities.

⚡ Backend — Python and FastAPI

FastAPI handles application requests and coordinates the services responsible for image assessment, speech transcription, agricultural knowledge retrieval, and environmental data processing.

🌿 Computer Vision — DINOv2 and ONNX Runtime

The crop-assessment service uses a DINOv2-based ONNX classification pipeline with 387 output classes.

Inference is performed through backend services when the required model is available. The system is experimental and requires further real-world accuracy validation.

🎙️ Speech Recognition — Faster-Whisper

Faster-Whisper-Tiny enables backend speech-to-text processing for multilingual agricultural interactions.

This helps users interact with the platform through spoken questions rather than depending