India has over 12 million gig delivery workers. When it rains too hard,
they stop earning. No insurance covers this. We are building one.
This is what we learned building RiskShield-Gig in Phase 2 of
Guidewire DEVTrails 2026.
What Is RiskShield-Gig?
A parametric insurance platform that pays gig delivery workers
automatically when external disruptions hit their zone - rain, floods,
curfews. No claim form. No waiting. The payout arrives before the
worker even knows a claim was processed.
Workers pay a weekly premium (₹20-₹40 based on zone risk). When a
trigger fires, money hits their UPI wallet in minutes.
Mekala-Sanjith3
/
RiskShield-Gig
AI-powered parametric insurance platform for gig delivery workers. Automatically detects disruptions (rain, floods, curfews) and pays out income loss to Swiggy/Zomato partners instantly - no claims, no forms, no waiting.
RiskShield-Gig 🛡️
AI-Powered Parametric Insurance for Gig Delivery Workers
Guidewire DEVTrails 2026 | Team: Prime AutoBots
📌 The Problem
India has over 12 million gig delivery workers. Most of them earn between ₹10,000 and ₹15,000 a month - roughly ₹600 to ₹800 a day - working for platforms like Swiggy and Zomato. When a heavy rainstorm hits, when a city-wide curfew is announced, or when flooding shuts down entire zones, these workers simply stop earning. There is no compensation. No claim to file. No safety net.
A worker in Hyderabad during the 2024 monsoon season could lose 8 to 10 working days. That is anywhere between ₹4,800 and ₹8,000 gone - with no recourse. Traditional insurance products do not address this. They are too expensive, too complex, and built for a workforce that earns a salary, not a daily wage.
RiskShield-Gig exists to close that gap.
💡 Proposed Solution
RiskShield-Gig…
What We Shipped in Phase 2
1. Worker Registration and AI Risk Profiling
Onboarding collects phone number, Aadhaar ID, delivery platform ID,
and UPI handle. The moment a worker registers, our Random Forest model
scores their zone using:
- Historical rainfall and flood data
- Average AQI over the past 6 months
- Delivery activity density in their area
This produces a weekly premium tier - Low (₹20), Medium (₹30),
or High (₹40).
The key decision: hyper-local pricing over city-level pricing.
A worker in Kondapur and a worker in LB Nagar face completely different
flood risks even though both are in Hyderabad. City-level pricing
overcharges one and undercharges the other.
2. Dynamic Premium Calculation
The model recalculates premiums weekly as new data comes in. Workers
in zones with a clean week see their score nudge down. Zones that had
disruptions nudge up. Premiums are capped at 5% of average weekly
income to stay affordable.
3. Automated Parametric Triggers
We integrated 4 live triggers:
| Trigger | Threshold | API |
|---|---|---|
| Heavy rainfall | Above 50mm/hr | OpenWeatherMap |
| Severe pollution | AQI above 350 | OpenAQ |
| Flood alert | Red alert issued | IMD mock |
| Curfew | Zone shutdown confirmed | Government mock |
When any trigger fires in an active subscriber's zone — the claim
pipeline starts. Automatically. The worker does nothing.
4. Claims Management
Policy creation is fully automated. Workers never "apply" for coverage.
The system creates and manages their policy from the moment they
subscribe.
Workers can view:
- Active coverage period (Monday to Sunday)
- Which triggers are active in their zone right now
- Full payout history
The Hardest Problem: GPS Spoofing
The biggest challenge was not building the happy path. It was
protecting against organized fraud rings using GPS spoofing apps to
fake their location inside a disruption zone while sitting safely at home.
A syndicate of 500 workers doing this simultaneously can drain a
liquidity pool in hours.
Our solution: GPS accounts for less than 20% of the fraud risk score.
The remaining weight comes from signals a spoofer cannot easily fake:
- Movement patterns - real bike speed vs static position
- Device sensors - accelerometer confirms physical motion
- Cell tower triangulation - cross-validated against GPS coordinates
- Platform activity - active orders vs zero activity during claim
- Crowd intelligence - are all nearby workers behaving consistently?
We use Isolation Forest for anomaly detection to catch:
- Sudden location jumps (5km+ in under 1 minute)
- Synchronized claim spikes from the same geo-cluster
- First-time accounts claiming only during major disruption events
The Fairness Problem
A genuine worker in heavy rain has poor GPS signal and slow network
connectivity - which looks identical to spoofing on the surface.
Binary approve/reject would hurt honest workers. We designed a
tiered response instead:
| Fraud Score | Response |
|---|---|
| 0 – 40 | Instant payout, no action needed |
| 41 – 70 | OTP + selfie (60 seconds) |
| 71 – 100 | Manual review with one-tap appeal |
Workers with long clean claim histories get extra trust weight. A worker
with 12 months of clean claims is treated very differently from an
account created 3 days ago.
What We Learned
Parametric insurance lives or dies on its fraud layer. Automated
payouts without robust validation is just a money tap. Fraud
architecture is not a feature - it is the foundation.
Design for zero patience. Every extra step is a drop-off for this
user base. Onboarding under 3 minutes. Zero worker action for claims.
If it is not instant, it does not work.
Weekly pricing builds trust. Workers who see a small predictable
₹20–₹40 deduction every Monday stay subscribed far longer than those
asked for a monthly lump sum.
What Is Next
Phase 3 focuses on advanced GPS spoofing defense in production,
instant payouts via Razorpay test mode, and intelligent dashboards
for workers and insurers.
Final DemoJam at DevSummit 2026 is the goal.
Team Prime AutoBots - Guidewire DEVTrails 2026
GitHub: https://github.com/Mekala-Sanjith3/RiskShield-Gig
Top comments (0)