AgroStack 🌴

 █████╗  ██████╗ ██████╗  ██████╗ ███████╗████████╗ █████╗  ██████╗ ██╗  ██╗
██╔══██╗██╔════╝ ██╔══██╗██╔═══██╗██╔════╝╚══██╔══╝██╔══██╗██╔════╝ ██║ ██╔╝
███████║██║  ███╗██████╔╝██║   ██║███████╗   ██║   ███████║██║      █████╔╝ 
██╔══██║██║   ██║██╔══██╗██║   ██║╚════██║   ██║   ██╔══██║██║      ██╔═██╗ 
██║  ██║╚██████╔╝██║  ██║╚██████╔╝███████║   ██║   ██║  ██║╚██████╗ ██║  ██╗
╚═╝  ╚═╝ ╚═════╝ ╚═╝  ╚═╝ ╚═════╝ ╚══════╝   ╚═╝   ╚═╝  ╚═╝ ╚═════╝ ╚═╝  ╚═╝

🏆 1st Prize Winner — IEDC Cluster Level Agricultural Hackathon

An award-winning, weather-aware Agronomic Expert System & AI Price Engine optimized for Kottayam District, Kerala. AgroStack empowers local farmers by bridging advanced predictive finance, real-time agronomic risk analysis, simulated federated learning, and accessible voice-based IVR systems.

---

🚀 Key Innovation Pillars

1. Hybrid AI Price Engine & XAI

• Prophet + LSTM Fusion: Blends a Meta Prophet seasonality model (70% weight) for long-term trends and a TensorFlow LSTM residual model (30% weight) to capture short-term price shocks. $$\text{Predicted Price} = (\text{Prophet Trend} \times 0.7 + \text{LSTM Residual} \times 0.3) \times \text{Advisory Bias}$$
• Explainable AI (XAI): Quantifies feature attributions, explaining price fluctuations (e.g., rainfall deficits, market demand volatility) in plain language.
• Cryptographic Integrity: Predictions are dynamically signed in-memory using RSA-PSS (2048-bit / SHA-256), creating an immutable audit trail for smart-contracts.

2. Weather-Aware Agronomic Advisory

• 25-Crop Knowledge Base: Evaluates live weather against biological thresholds for Kerala crops (Rubber, Black Pepper, Cardamom, Paddy, Coconut, etc.).
• Dynamic Supply Multipliers: Triggers supply-shock alerts and bias modifiers if thresholds are breached (e.g., wind speeds $>40\text{ km/h}$ trigger banana lodging alerts; humidity $>85%$ triggers pepper fungal rot risk).

3. Accessible Multi-Channel Communication (Twilio IVR & SMS)

• Malayalam Voice Portal: Farmers without active internet access can manage orders through an automated Twilio Interactive Voice Response (IVR) system.
• On-the-Fly Speech Synthesis:
  • Farmer presses 3 $\rightarrow$ Fetches inventory from Supabase.
  • Summarizes stock metrics using Groq LLM in Malayalam.
  • Generates Malayalam audio dynamically via Murf AI (Alicia voice) and streams it back to the call in real time.
• SMS & Emails: Dispatches transaction alerts via automated SMS and templates built with React Email & Nodemailer.

4. Privacy-First Simulated Federated Learning

• Regional Edge Training: Trains regional LSTM models (Kottayam, Idukki, Ernakulam) on local price datasets without centralizing raw financial data.
• FedAvg Optimization: Implements decentralized federated averaging to build a robust global forecasting model.
• Optimal Routing Recommendations: Computes 7-day average predictions to direct farmers to the highest-yielding regional market (e.g., advising a farmer in Kottayam to sell rubber in Idukki for higher returns).

5. What-If Profit Simulator

• A deterministic calculator analyzing land size, labor cost, and fertilizer cost against live mandi prices (from Data.gov.in) and current weather conditions (from Open-Meteo) to guide harvest timing and storage decisions.

6. Sarvam AI & Supabase MCP JSON-RPC Server

• An MCP-compatible API server exposing secure data-access tools (push_to_inventory and pull_from_inventory) integrated with Sarvam AI's multilingual LLM calling layer.

---

🛠️ The Tech Stack

Web & API Framework

• Frontend: Next.js 15 (App Router), React 19, TypeScript
• State & Communication: tRPC (Client & Server) for type-safe endpoints, TanStack React Query
• Styling & Motion: Tailwind CSS 4, Radix UI, Framer Motion (motion/react)
• Authentication: Better Auth (with Drizzle Adapter & cookie security)
• Database: Supabase (Postgres), Drizzle ORM

AI / Data Engineering

• Machine Learning: Meta Prophet (cyclical trends), TensorFlow Keras (LSTM sequential modelling & federated simulations)
• LLM & Speech: Groq SDK (gpt-oss-120b), Murf AI API, Sarvam AI Agent
• Background Jobs: Inngest (handles cron tasks like daily automated Facebook page agricultural reports)
• APIs: Twilio (SMS & Voice), Data.gov.in (Agmarknet live wholesale price records), Open-Meteo (Real-time weather)
• Object Storage: Cloudflare R2 (S3-compatible asset bucket)

---

📁 Repository Structure

├── AiPriceEngine/             # Python ML Predictor & Simulation Backend
│   ├── engine.py              # Prophet + LSTM Hybrid model & Agronomic Advisory Layer
│   ├── federatedlearning.py   # Simulated Federated Learning & FedAvg pipeline
│   ├── simulation_engine.py   # Farmer What-If simulator (Live Agmarknet + Open-Meteo)
│   ├── sarvamXsupa.py         # MCP JSON-RPC 2.0 server with Sarvam Agent tools
│   └── main.py                # FastAPI server hosting signed prediction endpoints
├── app/                       # Next.js App Router (Dashboard, Farmer & Buyer portals)
│   ├── api/                   # API Routes (tRPC, twilio, stts, auth callbacks)
│   └── page.tsx               # High-fidelity landing page
├── components/                # Shared React UI components
├── db/                        # Drizzle schema definition & migrations
├── inngest/                   # Inngest configurations (daily cron Facebook posters)
├── lib/                       # Server-side helper modules (Twilio, S3/R2, Mailer)
└── trpc/                      # tRPC routers (Inventory, Listings, Orders, Messages)

---

⚙️ Quick Start

1. Configure Environments

Create a .env file at the root:

cp .env.example .env
# Fill in your database, Twilio, Murf, and Groq keys.

2. Launch the AI engine (Python backend)

cd AiPriceEngine
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --host 0.0.0.0 --port 8000 --reload

3. Launch the Web App

# From repository root
bun install   # or npm install
bun dev       # Next.js local development server

---

AgroStack: Revolutionizing agriculture for Kerala’s farmers, one call and prediction at a time.