DharaniAI：為印度旱地農民提供離線氣候韌性及母語語音指引

DharaniAI (ధరణి / धरणी): Offline Climate Resilience & Native Voice Advisory for Indian Drylands 🌾☀️🔊

An Offline Edge AI Decision Engine for Smallholder Farmers in Extreme Drought Zones, Powered by Google Gemma 2.

Built to combat climate change, drought vulnerability, and rural connectivity deserts.
Targeting: Open-Source AI & Gemma 2 edge solutions.

🌍 The Problem: Climate Crisis in India's Drylands

Over 68% of India's cultivable agricultural land is chronically vulnerable to drought. Semi-arid rain-shadow regions like:

Rayalaseema (Anantapur / Kurnool, Andhra Pradesh)
Marathwada & Vidarbha (Latur, Beed, Maharashtra)
Bundelkhand (Jhansi, Mahoba, UP/MP border)
North Karnataka (Vijayapura, Bagalkot)
Western Rajasthan (Thar Desert border)

...face devastating crop loss due to recurring heatwaves (42°C+), shifting monsoons, and extreme evapotranspiration.

Why Conventional Solutions Fail Farmers:

Connectivity Deserts: During intense heatwaves or storm-triggered grid failures, cellular towers frequently lose power. Cloud-dependent AI apps become completely inaccessible when farmers need urgent guidance most.
Raw Forecasts vs Actionable Science: A standard weather app reporting "41°C and 25% humidity" does not tell a farmer how many liters of water their soil is losing per hour, or how to prevent flower drop in groundnut/cotton.
The Literacy & Language Barrier: Dense technical PDF advisories issued by universities are difficult for grassroots farmers to interpret.

💡 The Solution: DharaniAI (ధరణి / धरणी)

DharaniAI combines real micro-climate physics, official ICAR & CRIDA (Central Research Institute for Dryland Agriculture) contingency plans, and Google's open-weight Gemma 2 running 100% locally and offline on edge devices.

Most importantly, it features Native Audio Speech Synthesis, speaking directly to farmers in their mother tongue (Telugu, Hindi, Marathi, Kannada, English) through laptop or smartphone speakers!

🔬 Core System Architecture

┌────────────────────────────────────────────────────────────────────────┐
│                   DHARANI AI (Offline Edge System)                     │
├────────────────────────────────────────────────────────────────────────┤
│ 1. Micro-Climate Physics Engine                                        │
│    -> Tetens Equation: Saturation vs Actual Vapor Pressure             │
│    -> Vapor Pressure Deficit (VPD in kPa)                              │
│    -> FAO-56 Hargreaves-Samani: Evapotranspiration (ET0 in L/ha/day)   │
│                                                                        │
│ 2. Indian Dryland Agronomy Knowledge Base (RAG)                        │
│    -> ICAR / CRIDA Drought Contingency Frameworks                     │
│    -> Critical crop water sensitivity stages                           │
│    -> Organic foliar anti-transpirants (Kaolin, KNO3, Jeevamrutha)    │
│                                                                        │
│ 3. Gemma 2 Local Intelligence (via Ollama)                             │
│    -> General reasoning across thousands of crops & pest symptoms      │
│    -> Translates complex agro-physics into empathetic action plans     │
│                                                                        │
│ 4. Native Voice & Regional Audio Synthesizer                           │
│    -> Automatic voice audio generation in Telugu, Hindi, Marathi, etc. │
│    -> Speaks actionable advice aloud for grassroots farmers            │
└────────────────────────────────────────────────────────────────────────┘

✨ Key Capabilities

1. ☀️ Real-Time Agro-Meteorological Physics

Vapor Pressure Deficit (VPD): Measures the atmospheric drying power. When VPD > 2.5 kPa, plant stomata violently close, halting photosynthesis and causing flower drop.
Reference Evapotranspiration ($ET_0$): Calculates the exact water volume evaporating from the soil canopy (e.g. 7.5 mm/day = 75,000 Liters/hectare/day).

2. 🌾 Structured ICAR/CRIDA Crisis Strategies

Emergency Drought Action Plans: Specific anti-transpirant recipes (e.g. 1% Potassium Nitrate, 5% Kaolin clay) to reduce transpirational water loss and cool canopies by 2–3°C.
Post-Sunset Pulse Irrigation: Scheduling water application after 7 PM to prevent up to 60% surface evaporation loss.
Soil Sponge Building: Crop residue mulching, biochar, and broad-bed furrowing to maximize soil water retention.

3. 🔊 Native Speech for Farmers

Farmers don't need to read walls of text. DharaniAI generates a crystal-clear, warm voice audio briefing in:

Telugu (తెలుగు)
Hindi (हिन्दी)
Marathi (मराठी)
Kannada (ಕನ್ನಡ)
English

4. 🧠 Beyond Presets: General Crop & Climate Intelligence

While pre-loaded with major dryland crops (Groundnut, Cotton, Soybean, Pearl Millet, Pigeonpea), Gemma 2 can evaluate any custom observation or question entered by the farmer:

"My tomato leaves are curling under high heat and whiteflies are swarming..."
"Borewell water has high salinity (EC 3.2), how should I irrigate my chillies?"

🔮 Why Open-Source AI & Gemma 2 are Essential

Edge Deployment in Remote Villages: Google's Gemma 2 (gemma2:2b) delivers high-tier scientific reasoning while running smoothly on modest local hardware (even a single laptop or edge gateway) with zero internet access.
Data Privacy & Zero Cloud Subscriptions: Village panchayats, cooperatives, and smallholders cannot afford recurring per-token cloud API bills. DharaniAI operates at zero operational marginal cost.

💻 Open-Source Code & Repository

GitHub Repository: https://github.com/volety17/ghost-of-lords (see /dharani_ai)
Tech Stack: Python 3.11, Google Gemma 2, Streamlit, Ollama, gTTS / pyttsx3, FAO-56 Agro-Meteorology algorithms.

🚀 Running DharaniAI Locally

git clone https://github.com/volety17/ghost-of-lords.git
cd ghost-of-lords
pip install streamlit gTTS pyttsx3 requests
ollama pull gemma2:2b
python -m streamlit run dharani_ai/app.py

Dedicated to the resilient farmers of India's drylands. 🌾🇮🇳