← Index Case study 02

AgriVision AI

Crop monitoring platform with AI-driven disease detection

Stack
ReactTypeScriptPythonPyTorchVite
Links
Source ↗ Live demo ↗
fig. — simulated system behaviour live sim
Problem

Farmers need a way to monitor crop health across fields, receive timely alerts, and identify plant diseases early — without requiring agricultural expertise.

Solution

A full-stack platform where farmers register fields, upload crop images for AI analysis, track weather conditions, and receive disease alerts with treatment recommendations.

Architecture

Frontend

  • React 18 + TypeScript + Vite SPA
  • Pages: Landing, Auth, Fields, Analysis, Alerts, Analytics, Settings
  • UI: Radix UI + TailwindCSS + Framer Motion
  • 3D: React Three Fiber (plant visualization)
  • Data: TanStack Query + React Hook Form + Zod

Backend

  • Node.js + Express
  • Auth: JWT-based authentication
  • Database: SQLite with FK constraints
  • Files: Upload + automatic thumbnail generation
  • AI: Python/PyTorch integration (disease detection pipeline)
  • Jobs: Background analysis processing
Key Decisions

SQLite for early-stage deployment

No DB server needed — simplifies local dev and initial hosting without sacrificing relational integrity.

PyTorch + Torchvision for disease classification

Pre-trained models with transfer learning for accurate plant disease identification from uploaded images.

Three.js for immersive landing page

Engagement signal, not decoration — a 3D plant visualization that communicates the product's domain instantly.

Firebase Admin SDK for push notifications and remote config

Real-time disease alerts delivered to farmers' devices with remotely configurable notification thresholds.
Stack
Frontend

React, TypeScript, Vite, TailwindCSS, Radix UI, Three.js
Backend

Node.js, Express, SQLite, JWT, Firebase Admin SDK
AI

Python, PyTorch, Torchvision
Infrastructure

Vercel, Docker

Next case study

Mailing Service

→