KisanRakshak — Farmer Distress Early Warning System

SDG Alignment

• SDG 2: Zero Hunger — Prevents farmer distress by enabling early intervention
• SDG 3: Good Health and Well-being — Reduces farmer suicides through mental health alerts

Problem

Farmer distress in Vidarbha, Maharashtra leads to debt cycles, crop failures, and rising suicide rates. Authorities lack a real-time system to identify at-risk districts and deploy preventive measures.

Solution

An AI-powered early warning system that:

• Computes a Crop Distress Index (CDI) using 6 features (price volatility, rainfall, NDVI, groundwater, debt, historical base rate)
• Generates PDF distress reports with SHAP explainability for authorities
• Displays a real-time heatmap of all 10 Vidarbha districts
• Provides history replay via daily snapshots

Tech Stack

Layer	Technology
Frontend	React 18, Vite, TypeScript, Leaflet, Recharts
Backend (ML)	Django 5, DRF, scikit-learn, XGBoost, SHAP
Ingestion	Go 1.22 (worker pool, cron, WebSocket)
Database	PostgreSQL / SQLite
PDF	WeasyPrint

Architecture

5-Layer System Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                 LAYER 1 — REACT DASHBOARD (Port 3000)                    │
│                                                                         │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐  ┌────────────┐ │
│  │   MapView    │  │DistrictDetail│  │  CDITrend    │  │AlertPanel  │ │
│  │  (Leaflet)   │  │(SHAP bar)    │  │  (Recharts)  │  │(Realtime)  │ │
│  └──────┬───────┘  └──────┬───────┘  └──────┬───────┘  └─────┬──────┘ │
│  ┌──────┴───────┐  ┌──────┴───────┐  ┌──────┴───────┐  ┌─────┴──────┐ │
│  │ Notifications│  │   History    │  │   Header     │  │   Layout   │ │
│  │ (PDF mgmt)   │  │(snap replay) │  │(logo, WS)    │  │(tabs,dark) │ │
│  └──────────────┘  └──────────────┘  └──────────────┘  └────────────┘ │
└───────────────────────────────────┬─────────────────────────────────────┘
                                    │
                        ┌───────────┴───────────┐
                        │   REST API (JSON)      │
                        │   /api/v1/cdi/*        │
                        │   /api/v1/notifications │
                        │   /api/v1/snapshots/*  │
                        └───────────┬───────────┘
                                    ▼
┌─────────────────────────────────────────────────────────────────────────┐
│                  LAYER 2 — DJANGO ML API (Port 8000)                    │
│                                                                         │
│  ┌────────────────────┐  ┌────────────────────┐  ┌───────────────────┐ │
│  │    CDI Engine      │  │  SHAP Explainer    │  │  Feature Pipeline │ │
│  │  Ensemble Model    │  │  Feature Contrib.  │  │  Price Volatility │ │
│  │  RF + XGBoost + LR │  │  Waterfall Chart   │  │  Rainfall Dev.    │ │
│  └─────────┬──────────┘  └─────────┬──────────┘  │  NDVI Anomaly     │ │
│  ┌─────────┴──────────┐  ┌─────────┴──────────┐  │  Groundwater Dep. │ │
│  │  Notification Sys. │  │  Snapshot Engine   │  │  Debt Exposure    │ │
│  │  PDF (WeasyPrint)  │  │  Daily JSON State  │  │  Historical Base  │ │
│  └─────────┬──────────┘  └─────────┬──────────┘  └───────────────────┘ │
│  ┌─────────┴──────────┐  ┌─────────┴──────────┐                         │
│  │  DRF REST Views    │  │  Celery Workers    │                         │
│  │  7 API Endpoints   │  │  (async inference) │                         │
│  └────────────────────┘  └────────────────────┘                         │
└───────────────────────────────────┬─────────────────────────────────────┘
                                    │
                        ┌───────────┴───────────┐
                        │   Django ORM / SQL    │
                        └───────────┬───────────┘
                                    ▼
┌─────────────────────────────────────────────────────────────────────────┐
│               LAYER 3 — POSTGRESQL + TIMESCALEDB (Port 5432)            │
│                                                                         │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐  ┌────────────┐ │
│  │ Crop Prices  │  │ Weather Data │  │ Suicide      │  │ Groundwater│ │
│  │ (hypertable) │  │ (hypertable) │  │ History (dim)│  │ (hypertable)│ │
│  ├──────────────┤  ├──────────────┤  ├──────────────┤  ├────────────┤ │
│  │NDVI Index    │  │Loan Default  │  │ CDI Scores   │  │ Alerts     │ │
│  │(hypertable)  │  │(hypertable)  │  │ (hypertable) │  │ (hypertable)│ │
│  └──────────────┘  └──────────────┘  └──────────────┘  └────────────┘ │
│  TimescaleDB: auto-partitioned by time, continuous aggregates,          │
│  compression policies for efficient storage                             │
└───────────────────────────────────┬─────────────────────────────────────┘
                                    │
                        ┌───────────┴───────────┐
                        │   Read by ML API      │
                        │   Written by Go Ingest│
                        └───────────┬───────────┘
                                    ▼
┌─────────────────────────────────────────────────────────────────────────┐
│                  LAYER 4 — GO INGESTION (Port 8080)                     │
│                                                                         │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌────────────────────────┐ │
│  │AgMarknet │  │  IMD API │  │  NASA    │  │ India Data Portal     │ │
│  │ Fetcher  │  │ Fetcher  │  │ MODIS    │  │ Loan Default Fetch    │ │
│  │          │  │          │  │ Fetcher  │  │                       │ │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘  └───────────┬───────────┘ │
│       │             │             │                     │              │
│       └─────────────┴─────────────┴─────────────────────┘              │
│                      │  5-goroutine worker pool                        │
│              ┌───────┴───────┐                                         │
│              │  Go Router    │  (Chi/Gin) + Cron Scheduler             │
│              │  (6h/24h)     │  + WebSocket Broadcaster                │
│              └───────┬───────┘                                         │
└──────────────────────┼─────────────────────────────────────────────────┘
                       │
           ┌───────────┴───────────┐
           │  Write to TimescaleDB │
           │  Push alerts via WS   │
           └───────────────────────┘
                       ▼
┌─────────────────────────────────────────────────────────────────────────┐
│              LAYER 5 — ALERTING & NOTIFICATION                          │
│                                                                         │
│  CDI Thresholds:                                                        │
│  Critical (>=70)  → DBT Transfer + Loan Moratorium + SMS               │
│  High (50-69)     → Dashboard Alert + Email + This-week Visit          │
│  Medium (30-49)   → Dashboard Highlight + Monitoring                   │
│  Low (<30)        → Routine Monitoring                                 │
│                                                                         │
│  • WebSocket push → Dashboard real-time alert panel                    │
│  • PDF report generation → A4 report with SHAP + trend + actions       │
│  • Daily snapshots → JSON state capture for history replay             │
│  • Bhashini API → Hindi/Marathi alert text generation                  │
└─────────────────────────────────────────────────────────────────────────┘

Data Flow

1. Data Ingestion — Go service fetches govt APIs (AgMarknet, IMD, NASA) via 5-goroutine worker pool on cron schedule
2. CDI Computation — Django ML API fetches metrics, runs ensemble model (RF+XGB+LR), computes SHAP values, stores CDIRecords
3. Dashboard Visualization — React polls REST API, renders Leaflet heatmap, displays SHAP breakdown
4. Notification Pipeline — Click "Send" → Django renders A4 PDF via WeasyPrint → saves to MEDIA_ROOT → creates Notification record
5. History Snapshots — Daily snapshots serialize full CDI + alerts state → stored in CDISnapshot → replay any date
6. Real-time Alerts — Go polls every 15s → WebSocket push → Dashboard AlertPanel updates live

Dual-Stack Design

Concern	Go (ingest/)	Python (ml/)
Data Fetching	5 goroutines, 15s polling	N/A
ML / AI	N/A	scikit-learn, XGBoost, SHAP
PDF Generation	N/A	WeasyPrint (Django template)
Concurrency	Native goroutines + channels	Celery (optional)
Scheduling	Cron (native Go)	APScheduler / Celery Beat

Quick Start

# ML API
cd ml
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python manage.py migrate
python manage.py runserver 0.0.0.0:8000 --noreload

# Dashboard
cd dashboard
npm install
npm run dev

API Endpoints

• GET /api/v1/cdi/rankings — All districts ranked by CDI
• GET /api/v1/notifications — List notification history
• POST /api/v1/notifications/generate — Generate distress PDF
• GET /api/v1/snapshots — List history snapshots
• POST /api/v1/snapshots/generate — Create daily snapshot

Submission

This project was built as the Artificial Intelligence Final Project for the course Build a Solution for a Sustainable Development Goal (SDG).