Skip to content

RuthiKode/zerodown

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 

Repository files navigation

ZeroDown 🛡️

AI-Powered Parametric Income Insurance for India's Q-Commerce Delivery Partners

Zero income loss. Zero downtime. Zero paperwork.


The Problem

India's Q-Commerce delivery partners (Zepto, Blinkit, Swiggy Instamart) operate on brutal 10-minute SLAs. A single external disruption — heavy rain, a local curfew, elevated AQI — can wipe out an entire shift's earnings in minutes. These workers have no safety net. When disruptions hit, they bear 100% of the financial loss alone.

ZeroDown fixes that.


Our Persona

Segment: Grocery / Q-Commerce delivery partners (Zepto, Blinkit, Swiggy Instamart)

Why Q-Commerce specifically? Because a 10-minute delivery promise means any external disruption causes immediate, measurable income loss. The parametric trigger is tighter, the payout case is stronger, and the persona is underserved compared to food delivery workers.

Persona Scenarios

Scenario 1 — Ravi, Zepto partner, Bengaluru It's a Tuesday evening. Ravi completes 4 deliveries/hour earning ≈ ₹80/hour. At 6 PM, heavy rain (>65mm/hr) begins. Orders drop 80%. Ravi earns ₹0 for the next 3 hours. ZeroDown detects the weather trigger, auto-initiates a claim, and credits ₹200 to his UPI in under 2 minutes. No forms. No calls.

Scenario 2 — Priya, Blinkit partner, Delhi AQI in her zone crosses 400 (Severe). The platform deprioritises outdoor deliveries. Her active hours drop from 8 to 3. ZeroDown's parametric trigger fires, covering 60% of her projected income loss for the day automatically.

Scenario 3 — Arun, Swiggy Instamart partner, Mumbai A sudden local curfew is announced in his delivery zone. He cannot access pickup/drop locations for 5 hours. ZeroDown detects the zone-level social disruption event, cross-validates with his GPS inactivity, and processes the claim without him lifting a finger.


Application Workflow

Worker Onboards (Web + Mobile)
        │
        ▼
Risk Profile Generated (AI/ML)
        │
        ▼
Weekly Policy Purchased (₹25–₹60/week)
        │
        ▼
Real-Time Disruption Monitoring (APIs)
        │
        ▼
Parametric Trigger Fires
        │
        ▼
Auto Claim Initiated (Zero Touch)
        │
        ▼
Fraud Detection Check (AI)
        │
        ▼
Instant UPI Payout
        │
        ▼
Worker Notified (Push / SMS)

Weekly Premium Model

ZeroDown prices insurance on a weekly basis to align with the gig worker earnings cycle. Workers earn and spend week-to-week — so we insure week-to-week.

How Pricing Works

Risk Tier Weekly Premium Max Weekly Coverage Who Qualifies
Low Risk ₹25 ₹1,000 Zones with low historical disruption, high Trust Score
Medium Risk ₹38 ₹1,500 Zones with moderate weather/traffic disruption history
High Risk ₹55 ₹2,000 Flood-prone zones, high AQI history, strike-prone areas

Dynamic Pricing Factors (AI-Driven)

The weekly premium is not fixed — it adjusts each week based on:

  • Zone Risk Score — historical disruption frequency in the worker's primary delivery zone
  • Worker Trust Score — based on GPS consistency, claim history, delivery activity
  • Predictive Weather Forecast — upcoming week's weather risk (rainfall probability, AQI forecast)
  • Seasonal Modifiers — monsoon season, Diwali traffic, harvest festivals

Example: A Zepto partner in Koramangala (low flood history, high trust score) pays ₹28/week. The same worker during monsoon season (June–September) pays ₹42/week because the predictive model flags elevated rainfall risk.


Parametric Triggers

These are the 5 automated triggers that fire a claim — no manual filing required.

# Trigger Condition Data Source
1 Heavy Rainfall Rainfall > 65mm/hr in delivery zone OpenWeather API
2 Severe AQI AQI > 300 (Very Poor / Severe) in zone CPCB AQI API
3 Extreme Heat Temperature > 44°C AND humidity > 60% OpenWeather API
4 Zone Lockdown / Curfew Government-issued curfew or Section 144 in zone News API + manual override
5 Delivery Demand Collapse Platform order volume drops > 70% in zone for > 45 min Simulated platform API

Claim Validation Logic

A trigger alone doesn't release payout. The system cross-validates:

  1. External signal confirmed (API data)
  2. Worker GPS shows they were in the affected zone during the event
  3. Worker activity dropped (delivery count fell during the window)
  4. No duplicate claim for the same event window
  5. Fraud score below threshold

All five checks pass → auto-approve → instant payout.


AI / ML Integration Plan

1. Dynamic Premium Calculation (Random Forest)

  • Input features: Zone historical disruption rate, worker trust score, upcoming weather forecast, seasonal index, platform demand patterns
  • Output: Personalised weekly premium (₹)
  • Why Random Forest: Handles mixed feature types well, interpretable for regulators

2. Income Loss Prediction (Linear Regression + Zone Baseline)

  • Input: Worker's average hourly earnings × disruption duration × demand drop %
  • Output: Estimated income loss in ₹ (payout amount)
  • Calibrated per zone — a worker in high-density Koramangala has a different baseline than Whitefield

3. Fraud Detection (Isolation Forest + Rule Engine)

  • Signals monitored:
    • GPS location inconsistency (claimed to be in zone but wasn't)
    • Worker was already inactive before disruption (can't claim disruption caused inactivity)
    • Duplicate claim attempt for same event
    • Coordinated fraud (multiple workers filing for a non-existent event)
  • Output: Fraud probability score (0–1). Score > 0.75 = flagged for review

4. Worker Trust Score (Weighted Scoring Model)

  • Computed weekly. Factors: claim history accuracy, GPS data quality, delivery activity consistency
  • Higher trust → lower premium, faster payouts

5. Predictive Risk Alerts (Prophet / LSTM — Phase 3)

  • Forecasts zone disruption risk for the upcoming week
  • Powers the pre-week premium adjustment and sends risk alerts to workers

Platform Choice — Web + Mobile

Why both?

Platform Audience Key Use Case
Web (Next.js) Insurer admin, partner managers Dashboard, analytics, fraud review, policy management
Mobile (React Native / Flutter) Delivery workers Policy purchase, claim status, payout notifications

Workers interact on mobile — fast, thumb-friendly UI with minimal steps. Admins and insurers use the web dashboard for analytics and oversight. Both share the same FastAPI backend.


Tech Stack

Frontend

  • Web: Next.js 14, Tailwind CSS, ShadCN UI
  • Mobile: React Native (Expo) — targeting Android first (80%+ of delivery partners use Android)

Backend

  • Primary API: FastAPI (Python) — handles claims, triggers, fraud detection, AI models
  • Supporting services: Node.js — real-time WebSocket notifications, event streaming

Database

  • PostgreSQL — workers, policies, claims, payouts
  • Redis — session management, real-time event cache, trigger state
  • PostGIS extension — geo-spatial zone risk mapping

AI / ML

  • Scikit-learn — Random Forest (pricing), Isolation Forest (fraud)
  • Facebook Prophet — earnings forecasting, seasonal adjustments
  • Pandas / NumPy — feature engineering

Integrations

  • OpenWeather API (weather triggers)
  • CPCB AQI API (pollution triggers)
  • News API (social disruption signals)
  • Razorpay sandbox (mock UPI payouts)
  • Firebase Cloud Messaging (push notifications)

Infrastructure

  • Docker + Docker Compose for local dev
  • GitHub Actions for CI/CD
  • Render / Railway for deployment (free tier)

Development Plan

Phase 1 (Weeks 1–2) — Ideation & Foundation ✅

  • Persona definition and scenario research
  • Weekly premium model design
  • Parametric trigger specification
  • Tech stack decision
  • Basic project scaffolding (repo structure, CI setup)
  • Static UI wireframes (worker onboarding + dashboard)

Phase 2 (Weeks 3–4) — Automation & Protection

  • Worker registration and onboarding flow
  • Insurance policy creation and management
  • Dynamic premium calculation (ML model v1)
  • 3–5 parametric trigger integrations (mock APIs)
  • Basic claims management UI
  • Zero-touch claim initiation pipeline

Phase 3 (Weeks 5–6) — Scale & Optimise

  • Advanced fraud detection (GPS spoofing, duplicate claim detection)
  • Simulated instant payout via Razorpay sandbox
  • Intelligent dashboard — worker view + insurer admin view
  • Predictive analytics (next-week disruption risk)
  • Final demo video + pitch deck

Repository Structure

zerodown/
├── frontend/
│   ├── web/              # Next.js admin dashboard
│   └── mobile/           # React Native worker app
├── backend/
│   ├── api/              # FastAPI — core services
│   │   ├── claims/
│   │   ├── policies/
│   │   ├── triggers/
│   │   └── fraud/
│   ├── ml/               # ML models and feature pipelines
│   └── notifications/    # Node.js WebSocket + FCM service
├── data/
│   └── mock/             # Mock API responses for triggers
├── docker-compose.yml
└── README.md

Why ZeroDown Wins

  • Tighter persona: Q-Commerce's 10-minute SLA makes income loss from disruption immediate and measurable — cleaner payout logic than food or e-commerce
  • Zero-touch experience: Workers don't file claims. The system detects, validates, and pays. That's the product.
  • AI at every layer: Pricing, fraud, and forecasting are all ML-driven, not rule-based hacks
  • Weekly model is real: Aligns with how gig workers actually earn and plan — not monthly premiums they can't afford to front

Built for Guidewire DEVTrails 2026 — Unicorn Chase Team ZeroDown

About

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors