🌾 FarmAlert

మీ పొలం మా బాధ్యత — Your Farm. Stays Safe.

FarmAlert is a free, AI-powered agricultural decision-support platform built for smallholder farmers in Telangana, India. It delivers hyper-local climate risk forecasts, satellite-derived crop health analysis, bilingual AI advisories in Telugu and English, and a RAG-powered knowledge base grounded in official government documents — all in one place.

🌐 Live Demo: farmalert.vercel.app

---

🗺️ System Architecture

📄 Full Technical Documentation: FarmAlert_Technical_Documentation_v1.pdf

The system has two parallel AI pipelines that both converge at Gemini 2.0 Flash:

• Advisory pipeline — fetches live satellite + weather data, runs through risk and water engines, generates a bilingual action plan
• RAG knowledge base pipeline — embeds farmer questions, searches Pinecone vector index, retrieves grounded document chunks, generates verified answers

---

✨ Features

🌦️ Hyper-local Climate Advisory

• 14-day weather forecast via Open-Meteo API (temperature, rainfall, UV, ET0)
• 22-year climate baseline anomaly detection via NASA POWER API
• Real-time satellite indices via NASA MODIS — NDVI, EVI, LST, LAI (180-day history)
• 6 risk classifications — Heatwave, Drought, Flooding, Soil Moisture Drop, Wind Damage, Pest Risk
• FAO-56 Penman-Monteith crop water requirements with stage-specific Kc coefficients
• Bilingual daily / weekly / monthly action plans in Telugu + English

🤖 RAG Knowledge Base

• Floating chat widget on every page — glass pill design mirroring the navbar
• Answers grounded in 4 official government documents (346 vectors, Pinecone)
• Query enrichment with farm context (crop + stage + location) for precision retrieval
• In-memory OrderedDict cache (100 entries, LRU eviction) — ⚡ instant on repeat questions
• Animated letter-by-letter placeholder text, macOS-style expand animation
• Attach saved farms from My Farms to personalize answers

🗺️ Farm Management

• Interactive Leaflet map for GPS-based farm location
• Save and manage multiple farms with risk history
• My Farms dashboard with last risk status per farm

---

🏗️ Tech Stack

Layer	Technology
Frontend	React 18 · Vite · TailwindCSS · Leaflet · Framer Motion · motion/react
Backend	FastAPI · Python 3.11 · asyncio · uvicorn
AI — Advisory	Gemini 2.0 Flash (gemini-2.0-flash)
AI — RAG Generation	Gemini 2.5 Flash (gemini-2.5-flash)
Embeddings	gemini-embedding-001 · 768 dimensions · Matryoshka truncation
Vector Database	Pinecone Serverless · farmalert-kb · cosine similarity · AWS us-east-1
Weather API	Open-Meteo (14-day forecast · ET0)
Climate API	NASA POWER (22-year baseline)
Satellite API	NASA MODIS (NDVI · EVI · LST · LAI)
Frontend Hosting	Vercel
Backend Hosting	Render (free tier · 512MB RAM)

---

📚 Knowledge Base Documents

Document	Source	Pages	Chunks
PM-KISAN Operational Guidelines	pmkisan.gov.in	12	15
Telangana Kharif Agro-Advisory 2025	icar.org.in	7	12
ICAR Soil & Nutrient Management (NRM-2702)	icar.org.in	115	104
PMFBY Crop Insurance Guidelines	pmfby.gov.in	144	215
Total		278	346

---

🔌 API Endpoints

Method	Endpoint	Description
POST	/api/analyze	Full farm analysis — weather + satellite + risk + water + Gemini advisory
POST	/api/knowledge-base	RAG Q&A — embedding + Pinecone search + grounded Gemini answer
GET	/	Health check

/api/analyze Request

{
  "lat": 17.9784,
  "lng": 79.5941,
  "crop_type": "cotton",
  "crop_stage": "flowering",
  "soil_type": "black",
  "location_name": "Warangal, Telangana"
}

/api/knowledge-base Request

{
  "question": "Am I eligible for PM-KISAN scheme?",
  "crop": "cotton",
  "stage": "flowering",
  "location": "Warangal, Telangana"
}

/api/knowledge-base Response

{
  "answer": "To be eligible for PM-KISAN...",
  "cached": false
}

---

🚀 Local Setup

Prerequisites

• Python 3.11+
• Node.js 18+
• Pinecone account (free tier)
• Google AI Studio API key(s)

Backend

cd backend
python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt

Create backend/.env:

GEMINI_API_KEY=your_gemini_api_key
GEMINI_RAG_API_KEY=your_rag_gemini_api_key
PINECONE_API_KEY=your_pinecone_api_key
PINECONE_HOST=your_pinecone_index_host

python run.py

Frontend

cd frontend
npm install

Create frontend/.env:

VITE_API_URL=http://localhost:8000

npm run dev

Knowledge Base Ingestion

cd backend
# Add PDFs to data/source_docs/
python app/scripts/ingest.py

---

🏛️ Key Architecture Decisions

Decision	Reason
No LangChain	RAM constraints on Render free tier (512MB)
Raw Pinecone SDK	Lighter footprint, direct control
PyMuPDF for PDF extraction	pypdf and pdfplumber had silent text duplication bug on sliced PDFs
output_dimensionality=768	Matryoshka Representation Learning — officially supported truncation, not lossy
RETRIEVAL_DOCUMENT vs RETRIEVAL_QUERY	Different embedding modes for ingestion vs query time — improves retrieval quality
Two separate Gemini API keys	Isolates advisory and RAG quota pools — prevents one feature from exhausting the other
In-memory cache (no Redis)	Zero added dependencies, zero cost, adequate for current scale
localStorage for farms	No database yet — sufficient for MVP, clean migration path to Supabase later

---

📁 Project Structure

FARMALERT/
├── backend/
│   ├── app/
│   │   ├── main.py                 # FastAPI routes
│   │   ├── config.py               # Settings + env vars
│   │   ├── models.py               # Pydantic models
│   │   ├── services/
│   │   │   ├── gemini_service.py   # Advisory AI generation
│   │   │   ├── rag_service.py      # RAG pipeline
│   │   │   ├── risk_engine.py      # 6-type risk classification
│   │   │   ├── water.py            # FAO-56 water engine
│   │   │   ├── weather.py          # Open-Meteo integration
│   │   │   └── climate.py          # NASA POWER integration
│   │   └── scripts/
│   │       ├── ingest.py           # RAG document ingestion (local only)
│   │       └── verify_pinecone.py  # Index verification
│   ├── data/
│   │   └── source_docs/            # PDF knowledge base documents
│   └── requirements.txt
├── frontend/
│   └── src/
│       ├── pages/
│       │   ├── AnalysePage.jsx     # Main farm analysis
│       │   └── MyFarmsPage.jsx     # Saved farms
│       └── components/
│           └── KnowledgeChat.jsx   # RAG chat widget
└── docs/
    └── farmalert_complete_system_architecture.png

---

🌱 Roadmap

• Render deployment with all env vars (in progress)
• Frontend RAG chat UI connected to live backend
• Add 10–15 more knowledge base documents
• Supabase database for cross-device farm sync
• Migrate gemini_service.py to new google-genai SDK
• Mobile-responsive UI improvements
• Push notifications for critical risk alerts
• Telugu language input support in RAG chat

---

👨‍💻 Author

Suhas Naragani

• GitHub: @SuhasNaragani
• Project: farmalert.vercel.app

---

📄 License

This project is open source and available under the MIT License.

---

Built for Telangana farmers 🌾
మీ పొలం మా బాధ్యత