🌾 Smart Farming Advisor (Full‑Stack) — Agentic AI for Precision Agriculture

An end‑to‑end, agentic AI system that combines crop recommendation, plant disease detection, and farming Q&A — now with a modern Next.js + Tailwind frontend and a Flask API backend.

🚀 Double-click START.bat to launch the application!

View Demo

✨ Highlights

• 🌾 Crop Recommendation: Soil + climate‑aware crop prediction
• 🍃 Disease Detection: Image‑based plant disease classification
• ❓ Farming Q&A: RAG pipeline with semantic search + optional LLM
• 🧠 Agentic Routing: Smart intent routing between tools
• ⚡ Full‑Stack UI: Next.js 14 frontend with clean UX

---

📊 Performance Snapshot

Component	Model/Method	Score	Status
Crop Recommendation	Random Forest + Feature Engineering	99.39%	✅ Excellent
Disease Detection	ResNet50 (Transfer Learning)	98.97%	✅ Research‑grade
RAG Q&A	FAISS + Sentence Transformers	Hit Rate: 100%, MRR: 1.0	✅ Production‑ready
Agent Routing	Intent Classification	100%	✅ Perfect

---

📊 System Architecture

┌─────────────┐
│   Browser   │  ← User Interface
└──────┬──────┘
       │ HTTP
       ▼
┌──────────────────┐
│  Next.js (3000)  │  ← Frontend
│  - Pages         │
│  - Components    │
│  - API Client    │
└──────┬───────────┘
       │ REST API
       ▼
┌──────────────────┐
│  Flask (5000)    │  ← Backend
│  - Agent Router  │
│  - API Endpoints │
└──────┬───────────┘
       │
       ├──────────┬─────────────┬──────────┐
       ▼          ▼             ▼          │
  ┌────────┐ ┌─────────┐  ┌─────────┐    │
  │  Crop  │ │ Disease │  │   RAG   │    │
  │  Tool  │ │  Tool   │  │   Tool  │    │
  └────┬───┘ └────┬────┘  └────┬────┘    │
       │          │             │         │
       ▼          ▼             ▼         ▼
  ┌────────────────────────────────────────┐
  │          Trained ML Models             │
  │  - crop_model.pkl                      │
  │  - disease_model_resnet50.pth          │
  │  - faq_vector_store.index              │
  └────────────────────────────────────────┘

---

🧰 Tech Stack

Frontend

• Next.js 14 + TypeScript
• Tailwind CSS
• Axios / Fetch

Backend

• Python 3.8+
• Flask + Flask‑CORS
• PyTorch, Scikit‑learn
• FAISS + Sentence Transformers

---

🚀 Quick Start

Option 1: One-Click Start (Recommended)

Double-click: START.bat

This opens two terminals and starts both servers automatically!

Option 2: Manual Start

Terminal 1 - Backend:

cd "c:\Users\LAPTOP CLINIC\Documents\Projects\smart-farming-advisor"
python app.py

Terminal 2 - Frontend:

cd "c:\Users\LAPTOP CLINIC\Documents\Projects\smart-farming-advisor\frontend"
npm install  # First time only
npm run dev

Access: http://localhost:3000

---

🔗 API Endpoints

• GET /health - Health check
• POST /api/crop/recommend - Get crop recommendations
• POST /api/disease/predict - Detect plant disease (multipart form‑data with file)
• POST /api/qa - Ask farming questions
• GET /api/system/info - Get system and model information

---

⚙️ Environment Variables

Backend (optional):

GOOGLE_API_KEY=your_gemini_api_key  # For Gemini LLM (optional)

Frontend:

NEXT_PUBLIC_API_BASE=http://localhost:5000

---

📁 Project Structure

smart-farming-advisor/
├── app.py                       # Flask API
├── config.py                    # Configuration
├── requirements.txt             # Python dependencies
├── frontend/                    # Next.js app
│   ├── app/                     # Pages (home, crop, disease, qa, etc.)
│   ├── components/              # Reusable UI components
│   ├── lib/                     # API client & utilities
│   └── package.json             # Node dependencies
├── src/
│   ├── agent/                   # Agent router
│   ├── tools/                   # Crop / Disease / RAG tools
│   ├── models/                  # Training scripts
│   ├── preprocessing/           # Data preprocessing
│   └── evaluation/              # Evaluation scripts
├── models/                      # Trained models (large files ignored)
├── data/                        # Datasets (ignored)
├── outputs/                     # Results, plots (ignored)
└── notebooks/                   # Jupyter notebooks

---

🧪 Testing & Evaluation

Test the backend:

# Test agent
python src/agent/farming_agent.py

# Test individual tools
python src/tools/crop_predictor_tool.py
python src/tools/disease_detector_tool.py
python src/tools/rag_qa_tool.py

# Evaluate RAG system
python src/evaluation/rag_evaluation.py

Results are saved under outputs/.

---

🧠 Model Details

1) Crop Recommendation

• Algorithm: Random Forest Classifier
• Features: N, P, K, Temperature, Humidity, pH, Rainfall + engineered features
• Test Accuracy: 99.39%
• Training Time: ~2 seconds

2) Disease Detection

• Architecture: ResNet50 (Transfer Learning)
• Pretrained on: ImageNet
• Input: 224×224 RGB
• Test Accuracy: 98.97%
• Training Time: ~33 minutes (GPU)

3) RAG Q&A

• Embeddings: sentence-transformers/all-MiniLM-L6-v2
• Vector Store: FAISS
• Optional LLM: Google Gemini 2.0 Flash
• Performance: 100% hit rate, MRR 1.0

---

📚 Datasets

Dataset	Source	Purpose
Crop Recommendation	Kaggle	Soil‑based crop prediction (2,200 samples, 22 crops)
Plant Disease (PlantVillage)	Public	Disease classification (20,639 images, 15 diseases)
FAQ Knowledge Base	Custom	Farming Q&A (10 documents)

Large datasets and model binaries are excluded from GitHub. See .gitignore.

---

🛡️ GitHub & Large Files

This repo intentionally ignores:

• data/raw/ and data/processed/ - Raw and processed datasets
• Large model weights (*.pth, *.pt, *.onnx) - Trained model files
• FAISS index (*.index) - Vector store index
• Frontend build artifacts (frontend/.next, frontend/node_modules)
• Outputs and uploads (outputs/, uploads/)

Remember to add all these files by yourself, If you need any help running this project contact me at email.

---

🎯 Key Features

✅ Multi-modal AI System - Text, images, and structured data
✅ Agentic Routing - Intelligent query classification
✅ Transfer Learning - ResNet50 pretrained on ImageNet
✅ RAG Implementation - Semantic search with FAISS
✅ Production-Ready - 98-99% accuracy across all models
✅ Modular Architecture - Easy to extend and maintain
✅ Full-Stack - Modern frontend with Next.js + Tailwind
✅ Comprehensive Testing - All components tested

---

🤝 Contributing

Pull requests are welcome! Please open an issue for major changes.

---

📄 License

MIT License

---

🙌 Acknowledgments

• Datasets:

  • Crop Recommendation Dataset (Kaggle)
  • PlantVillage Disease Dataset
  • Farmer Support FAQ Dataset

• Pretrained Models:

  • ResNet50 (ImageNet, PyTorch)
  • sentence-transformers/all-MiniLM-L6-v2 (HuggingFace)

• Tools:

  • Google Colab (GPU training)
  • FAISS (Facebook AI Similarity Search)
  • Google Gemini API (LLM)

---

📬 Contact

Author: juni2003
Email: juni.xatti@gmail.com
GitHub: @juni2003

---

⭐ If you find this project helpful, please consider giving it a star!