Skip to content

thesantoshpant/cropscan

BranchesTags

Repository files navigation

CropScan

CropScan is a full-stack crop disease diagnosis app. A user signs in, uploads or captures a leaf photo, receives predictions from two PyTorch image classifiers, reviews AI-generated treatment guidance, and can ask follow-up questions about the diagnosis.

The app is built for smallholder farmers, backyard growers, and agricultural extension-style field work.

Features

  • Account signup, login, profile editing, and protected routes
  • Leaf image upload and browser camera capture
  • Dual-model prediction with EfficientNet-B0 and MobileNetV2
  • 38 PlantVillage disease and healthy classes
  • Low-confidence fallback behavior
  • Image suitability check before disease inference
  • Gemini-backed recommendation summary
  • Diagnosis follow-up chat with a 10-question limit per scan
  • Product-category and supply-plan recommendations
  • Saved scan history and field notes
  • Docker setup for one-command local runs
  • Vercel frontend and Render backend deployment path

Tech Stack

Layer Technology
Frontend React, TypeScript, Vite, Tailwind CSS, React Router
Backend FastAPI, Pydantic, PyMongo, JWT auth
AI / ML PyTorch, torchvision, EfficientNet-B0, MobileNetV2, Gemini
Database MongoDB Atlas or any MongoDB-compatible database
Local containers Docker Compose
Deployment Vercel frontend, Render backend

Repository Structure

backend/    FastAPI API, auth, MongoDB access, model loading, prediction, Gemini
cropscan/   React frontend
notebooks/  model training and experimentation notebooks

Required Accounts And Files

You need:

  • MongoDB connection string
  • JWT secret
  • Gemini API key if you want AI recommendations and chat
  • model files in backend/models

Required model files:

backend/models/efficientnet_b0_cropscan.pth
backend/models/mobilenetv2_cropscan.pth

If the model files are missing, the upload endpoint will fail when it tries to load the classifiers.

Environment Files

Use local .env files. Do not commit real secrets.

Create:

backend/.env
cropscan/.env

The repo includes:

backend/.env.example
cropscan/.env.example

Backend .env

ENVIRONMENT=development
MONGODB_URL=mongodb+srv://<username>:<password>@cluster.mongodb.net/cropscan
MONGODB_DB_NAME=cropscan
JWT_SECRET_KEY=replace-with-secrets-token-urlsafe-48-or-another-32-plus-character-secret
JWT_ALGORITHM=HS256
ACCESS_TOKEN_EXPIRE_MINUTES=60
PREDICTION_TOKEN_EXPIRE_MINUTES=30
CORS_ORIGINS=http://localhost:5173,http://127.0.0.1:5173
CORS_ORIGIN_REGEX=^https?://(localhost|127\.0\.0\.1)(:\d+)?$
MODEL_DIR=models
GEMINI_API_KEY=replace-with-your-gemini-api-key
GEMINI_MODEL=gemini-2.5-flash
RESEND_API_KEY=replace-with-your-resend-api-key
RESEND_FROM_EMAIL=CropScan <onboarding@yourdomain.com>
APP_BASE_URL=http://localhost:5173
PASSWORD_RESET_OTP_EXPIRE_MINUTES=10
EMAIL_DEBUG_OTP=false
RATE_LIMIT_ENABLED=true
RATE_LIMIT_STORAGE_URI=
TRUST_PROXY_HEADERS=false
PASSWORD_RESET_RESPONSE_DELAY_SECONDS=0.3
PASSWORD_RESET_MAX_ATTEMPTS=5
PASSWORD_RESET_LOCKOUT_MINUTES=60
MAX_SCANS_PER_USER=2000
MAX_PLOTS_PER_USER=50

For deployed production on Render, set:

ENVIRONMENT=production
CORS_ORIGINS=https://cropscan.tech,https://www.cropscan.tech
RATE_LIMIT_STORAGE_URI=rediss://<upstash-redis-url>
EMAIL_DEBUG_OTP=false

Use a shared Redis-compatible RATE_LIMIT_STORAGE_URI in production. Without it, rate limits are process-local and only correct for a single backend worker. Set TRUST_PROXY_HEADERS=true only behind a trusted proxy such as Render. Keep it false for local Docker or direct uvicorn runs.

Frontend .env

For normal local development:

VITE_API_BASE_URL=http://127.0.0.1:8000/api/v1

For Vercel production:

VITE_API_BASE_URL=/api/v1

Vercel uses cropscan/vercel.json to proxy /api/v1/* to the Render backend.

Fastest Local Setup With Docker

This is the easiest path for teammates and demo reviewers.

Windows PowerShell

Copy-Item backend\.env.example backend\.env
Copy-Item cropscan\.env.example cropscan\.env
docker compose up --build

macOS / Linux

cp backend/.env.example backend/.env
cp cropscan/.env.example cropscan/.env
docker compose up --build

Then open:

  • Frontend: http://localhost:5173
  • Backend: http://localhost:8000
  • Backend docs: http://localhost:8000/docs

Stop containers:

docker compose down

Docker still requires valid values in backend/.env.

Manual Local Setup

Use this path if you are developing without Docker.

Prerequisites

  • Python 3.11
  • Node.js 18+ and npm
  • MongoDB connection string

Python 3.11 is important because the backend uses PyTorch and torchvision.

Backend Setup

Windows PowerShell

cd backend
py -3.11 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
pip install -r requirements.txt
Copy-Item .env.example .env
uvicorn app.main:app --reload

macOS

If python3.11 is installed:

cd backend
python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install -r requirements.txt
cp .env.example .env
uvicorn app.main:app --reload

If you use Homebrew:

brew install python@3.11

Then rerun the setup above.

Linux

cd backend
python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install -r requirements.txt
cp .env.example .env
uvicorn app.main:app --reload

On Ubuntu/Debian, if Python 3.11 venv support is missing:

sudo apt update
sudo apt install python3.11 python3.11-venv python3.11-dev

Backend URLs:

  • API root: http://127.0.0.1:8000
  • Docs: http://127.0.0.1:8000/docs
  • Health: http://127.0.0.1:8000/health

Frontend Setup

Open a second terminal from the project root.

Windows PowerShell

cd cropscan
npm install
Copy-Item .env.example .env
npm run dev

macOS / Linux

cd cropscan
npm install
cp .env.example .env
npm run dev

Frontend URL:

http://127.0.0.1:5173

If Vite chooses another port, add that origin to backend/.env under CORS_ORIGINS or rely on the existing local CORS regex.

Using The App Locally

  1. Start the backend.
  2. Start the frontend.
  3. Open the frontend URL.
  4. Create an account or log in.
  5. Go to the scan page.
  6. Upload a leaf image or use camera capture.
  7. Review both model predictions.
  8. Review the recommendation and supply plan.
  9. Ask follow-up questions in chat.
  10. Open the dashboard to review saved history.

API Summary

Main routes:

  • GET /health
  • POST /api/v1/auth/signup
  • POST /api/v1/auth/login
  • GET /api/v1/auth/me
  • PATCH /api/v1/auth/me
  • POST /api/v1/auth/change-password
  • POST /api/v1/auth/forgot-password/request
  • POST /api/v1/auth/forgot-password/confirm
  • POST /api/v1/upload
  • POST /api/v1/chat

Protected routes require:

Authorization: Bearer <token>

Password reset uses a Resend-backed one-time code. In local development only, set EMAIL_DEBUG_OTP=true to return the code in the API response while email is not configured.

Verification Commands

Frontend

cd cropscan
npm run lint
npm run build

Backend

Windows PowerShell:

cd backend
.\.venv\Scripts\Activate.ps1
python -m pytest tests
python -m compileall app

macOS / Linux:

cd backend
source .venv/bin/activate
python -m pytest tests
python -m compileall app

Docker

docker compose build
docker compose up

Deployment

Production is split across Vercel and Render:

  • Frontend: Vercel
  • Backend: Render
  • Database: MongoDB Atlas
  • Domain: cropscan.tech

Render Backend

Create a Render Web Service from the repo.

Use:

Root Directory: backend
Environment: Docker
Branch: master
Health Check Path: /health

Set backend environment variables in the Render dashboard, not in Git.

Production CORS:

CORS_ORIGINS=https://cropscan.tech,https://www.cropscan.tech

Vercel Frontend

Create a Vercel project from the same repo.

Use:

Root Directory: cropscan
Framework Preset: Vite
Build Command: npm run build
Output Directory: dist
Production Branch: master

Set:

VITE_API_BASE_URL=/api/v1

The file cropscan/vercel.json forwards /api/v1/* requests to the Render backend and sends client-side routes back to the React app.

DNS

Point cropscan.tech and www.cropscan.tech to Vercel using the records shown in the Vercel dashboard. Use Vercel's displayed DNS values because they are the source of truth for the project.

Common Problems

py -3.11 or python3.11 is not found

Install Python 3.11 and recreate the backend virtual environment.

Upload fails when models load

Confirm both model files exist in backend/models.

Frontend cannot reach backend locally

Check:

  • backend is running
  • frontend .env uses http://127.0.0.1:8000/api/v1
  • backend .env includes the frontend origin or local CORS regex
  • backend was restarted after .env changes

Production CORS error

Check Render:

CORS_ORIGINS=https://cropscan.tech,https://www.cropscan.tech

Check Vercel:

VITE_API_BASE_URL=/api/v1

Then redeploy both services.

Render is slow on first request

Free Render services can sleep after inactivity. The first request may take longer while the service wakes and loads the PyTorch models.

Team Workflow

Use feature branches for active work, merge into develop after verification, then promote to master for deployment.

feature branch -> develop -> master

Run frontend and backend checks before merging into master.

About

cropscan-two.vercel.app

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages