🛡️ Shielded Rider

AI-Powered Monsoon Income Protection for Delivery Riders

---

Overview

Shielded Rider is a smart, automated parametric insurance system designed for two-wheeler delivery riders affected by monsoon disruptions.

Instead of manual claims and delayed payouts, our system:

• Predicts income loss

• Verifies real-world conditions

• Automatically compensates riders

---

Environment Variables & APIs

To run the application, you'll need to configure your environment variables.

1. Server (server/.env):
   • OPENWEATHER_API_KEY: Get a free key from OpenWeather for live weather risk scoring.
   • RAZORPAY_KEY_ID & RAZORPAY_KEY_SECRET: For simulating payouts (optional).
2. ML Service (ml-service/.env):
   • LLM_PROVIDER=openai
   • OPENAI_API_KEY: Your ChatGPT API Key (Required for ML reasoning).

---

How to Run via Docker

The platform is fully containerized using Docker Compose. Ensure you have Docker Desktop installed.

1. Clone the repository and navigate to the project directory.
2. Populate the .env files in server/ and ml-service/ as described above.
3. Run the following command from the root directory:

docker-compose up --build

4. Access the services:
   • Frontend App: http://localhost:5173
   • Backend Server: http://localhost:5000
   • ML Service: http://localhost:8001
   • MongoDB: Bound to port 27017

To run in the background, use -d flag. To stop the containers, run docker-compose down.

---

Problem

Delivery riders in Indian cities face:

• Heavy rainfall and flooding

• Unsafe road conditions

• Reduced order volume

This leads to 20–40% income loss, while traditional insurance:

• Is slow

• Requires manual claims

• Doesn’t cover daily earnings

---

Core Idea

• Predicts income disruption risk
• Tracks real-world conditions
• Estimates expected earnings
• Automatically compensates workers

No claims. No paperwork. Fully automated.

Key Innovation: Earnings Shadow Model

We don’t just detect rain — we calculate actual income loss.

How it works:

• Predict expected earnings

• Compare with actual earnings

• Trigger proportional payout

Example:

• Expected: ₹800

• Actual: ₹400

• Compensation: ₹400

---

Why This Matters

• Moves beyond event-based triggers → income-based protection
• Introduces real insurance logic (actuarial depth)
• Personalized per worker

---

System Workflow

1. Rider subscribes to a weekly plan

2. System collects:

   • Location data

   • Delivery activity

   • Weather data

3. Risk model predicts disruption

4. Earnings model estimates expected income

5. Fraud engine validates claim

6. Payout is triggered automatically

---

AI/ML Architecture

1. Risk Prediction Model

• Model: XGBoost

• Predicts probability of disruption

• Inputs:

  • Rainfall

  • Flood data

  • Traffic conditions

Output:

P(disruption) ∈ [0,1]

2. Earnings Prediction Model

• Model: LigthGBM

• Predicts expected daily income

• Inputs:

  • Historical earnings

  • Time & demand patterns

  • Weather

3. Fraud Detection Engine

• Model: Isolation Forest

• Detects anomalies and spoofing attempts

---

Anti-Fraud & Security System

Movement & Environmental Intelligence

• Matches rider movement with real traffic and weather conditions

• Detects unrealistic behavior

---

Cross-User Fraud Detection

• Identifies suspicious clusters of users with:

  • Same location

  • Same timestamps

  • Same inactivity

---

Active Movement Verification (Challenge-Response)

If GPS shows no movement during disruption:

• App prompts user to move ~100 meters

• Movement verified via:

  • Accelerometer

  • Gyroscope

  • GPS update

If verified: Claim proceeds
If not:
User is notified to check location services

Prevents fake “idle but claiming loss” scenarios

---

Transparent Anomaly Resolution

Instead of silent rejection:

• System notifies users when anomalies are detected

Example:
“Our system detected unusual device activity. Please reconnect or report the issue.”

Benefits:

• Builds trust

• Avoids unfair rejection

• Warns malicious users

---

Dynamic Pricing Model

Income Stability Score (ISS)

Premiums are personalized based on:

• Order consistency

• Ratings

• Active hours

• Work patterns

High ISS → Lower premium
Low ISS → Adjusted premium

Also considers:

• Local rainfall history

• Flood-prone zones

Risk Factor = (Local Rainfall History × Flood-Prone Zones)

ISS = (Avg Orders × Consistency × Rating × Active Hours × Risk Factor)

---

Policy Exclusions

No payouts in cases of:

• War or terrorism

• Government lockdowns

• Nuclear/biological events

• Proven fraud (GPS spoofing, tampering, account misuse)

---

Tech Stack

Layer	Technology
Frontend	React.js, Tailwind CSS, Vite
Backend	Node.js, Express
Database	MongoDB (Dockerized)
ML Svc	FastAPI, Scikit-learn, XGBoost
LLM	OpenAI (ChatGPT API)
APIs	OpenWeatherMap, OpenStreetMap
Payments	Razorpay (Test Mode)

---

Features Implemented

• Automated Weather Risk Scoring: Integrates with OpenWeather to assess real-time risk.
• Dynamic Payouts Simulation: Connects mock payments using Razorpay's API.
• LLM Reasoning Layer: Employs OpenAI's GPT models via a dedicated ML service to analyze claims and behavioral anomalies.
• Microservices Containerization: The entire application (Frontend, Backend, ML Service, and MongoDB) is fully dockerized for isolated, one-click deployments.

---

Impact

• Protects gig workers’ daily income

• Reduces financial uncertainty

• Builds trust with transparent systems

• Enables scalable, automated insurance

---

Future Scope

• Deep learning for advanced fraud detection

• Direct integration with Swiggy/Zomato APIs

• Hyperlocal flood prediction (500m grid)

---

Final Thought

RainShield Rider is not just an insurance system.

It is a financial safety net engineered for real-world gig workers, combining AI, behavioral analysis, and actuarial thinking to deliver fair, fast, and fraud-resistant protection.