We spent 15 days not writing code. Here's what we built instead.

Kotapothula Siva Raga Adithi — Fri, 20 Mar 2026 14:06:27 +0000

Day 1 of DEVTrails 2026 by Guidewire Software.

Four students from KL University. One problem statement.
Parametric insurance for gig workers.

Most teams open a problem statement and immediately think
about the tech stack.

We did that too — for about 10 minutes.

Then we stopped.

The Question That Changed Everything

We spent the next 15 days asking one question on repeat:

Who actually gets hurt when the storm hits?

The answer was always the same.

The Swiggy rider outside your window.
The Zomato partner sitting at home watching it rain,
knowing the day's earnings just disappeared.
The Zepto delivery partner who can't ride in a heatwave.

2.8 crore gig workers in India.
₹0 income protection built specifically for them.

Not health insurance. Not accident insurance.
Income insurance — the kind that triggers when it rains
and you simply cannot work.

That gap is what we built GigWeatherWage to close.

What is Parametric Insurance?

Traditional insurance works like this:
Something bad happens → you file a claim → a human reviews it
→ weeks later you maybe get paid.

Parametric insurance works differently.

You define the trigger condition upfront. The moment that
condition is met — rainfall above 50mm, AQI above 300,
a declared city curfew — the payout fires automatically.

No claim. No wait. No rejection.

For a gig worker who loses ₹800 on a rainy day, waiting
weeks for a manual claim review is not a solution.
Automatic payout in seconds is.

What GigWeatherWage Does

Workers register and choose a weekly plan:

Basic Shield — ₹20/week — ₹2,000 coverage
Storm Guard — ₹45/week — ₹3,500 coverage
Full Shield — ₹80/week — ₹5,000 coverage

The system monitors real-time weather APIs, AQI feeds,
and city alert systems for their registered zone.

When a threshold is crossed — payout goes directly
to their UPI account. Instantly.

Five disruption triggers:

Trigger	Threshold	Payout
Heavy Rainfall	> 50mm	₹300
Heatwave	> 42°C	₹250
Extreme AQI	> 300 AQI	₹200
City Curfew	Active alert	₹300
Platform Outage	> 2 hours	₹150

Each tied to a verifiable real-world data source.
No human decision required.

The Fraud Problem Nobody Talks About

On Day 12, Guidewire dropped a 24-hour challenge into
the hackathon.

A syndicate of 500 delivery workers had organized via
Telegram to use GPS spoofing apps — faking their locations
in rain zones while sitting safely at home — draining a
competitor's insurance pool.

Simple GPS verification was declared obsolete.

We had to redesign our fraud architecture under
extreme time pressure.

Our response — 5-signal AI fraud detection:

Signal 1 — Weather API Match (+40 pts)
Does it actually rain at the claimed location?
A spoofer can fake GPS. They cannot fake the weather
at that fake location.

Signal 2 — Movement Pattern (+30 pts)
A genuine stranded rider still moves — to shelter,
to a chai shop.
A spoofer sitting at home shows completely static
coordinates for hours.

Signal 3 — Device Integrity (+20 pts)
GPS spoofing apps run on emulators.
Emulators leave detectable signatures.

Signal 4 — Historical Behaviour (+15 pts)
Fraud rings create accounts fast.
A 3-day-old account claiming on the first major
weather event is suspicious.

Signal 5 — Geo-Cluster Detection (+30 pts)
47 workers all claiming from the exact same GPS
coordinates in the same 10-minute window is not
a coincidence.
That is the Telegram syndicate pattern.

Risk score decision:

Score	Decision
0 – 40	✅ Safe — instant payout
40 – 70	⚠️ Medium — 2hr verification window
70+	❌ Blocked — escalated for review

Genuine workers are never hard-blocked on a single signal.
That fairness layer was as important as the fraud
detection itself.

The 4 Personas — Live in the App

We built four interactive user personas that judges
and users can log into right now:

Persona	Type	Score	Outcome
👨 Raju Kumar	Genuine worker, 3 years	12	✅ Paid instantly
👩 Meena Devi	New account, genuine	55	⚠️ Delayed, then paid
👩 Priya Sharma	Real worker, insider fraud	85	❌ Blocked
🕵️ Vikram #7749	GPS spoof ring	135	❌ Blocked + ring flagged

Priya is the most important scenario.
She is a real registered Zomato worker — 8 months of
history, trusted device, looks completely legitimate.
But the weather API confirms no rain at her actual
device location.
The system catches her anyway.

The Feature Nobody Else Built

On Day 13 we had one realization:

A parametric insurance app that protects workers
during storms must work during storms — even when
the network is unreliable.

So we built offline-first claim filing.

Claims queue on-device when there's no network.
A fraud pre-check runs on cached data.
The moment connectivity returns — claim auto-syncs,
verification runs, payout processes.

The worker never loses their claim because of
bad signal in a storm.

No other solution addressed this.
For a product literally designed for storm conditions,
it felt like the most obvious thing in the world
once we saw it.

What We Actually Built

A fully deployed React prototype on Vercel with:

✅ Complete 5-step registration with UPI payment simulation
✅ 4 interactive fraud personas with live AI analysis
✅ 5-tab dashboard — Home, Claims, Payments, Alerts, Profile
✅ Offline detection with claim queuing and auto-sync
✅ Full claims history with fraud scores per claim
✅ UPI deep links that open PhonePe or GPay on real device

Tech stack:
React · Node.js · Python · MongoDB ·
OpenWeatherMap API · AQI API · Razorpay · Vercel

What We Learned

The hardest part of building any system is not
writing the code.

It is understanding the problem deeply enough that
the right solution becomes obvious.

We also learned:

Fraud detection is fundamentally a fairness problem, not a security problem
The GPS spoofing scenario taught us to design for adversarial users from Day 1
Offline-first is not optional for infrastructure that workers depend on during emergencies
Spending 15 days on the problem before touching a keyboard was the best decision we made

What's Next

Phase 2:
Python AI risk model · Live API integrations ·
Razorpay actual payouts · Flutter mobile app ·
Pilot with 10 real workers in Hyderabad

Phase 3:
City-wide disruption dashboards · Dynamic premium pricing ·
Platform API integrations · IRDAI parametric license ·
Expand to 5 Indian metros

We are not building for a hackathon.
We are building for the 2.8 crore workers who deserve
better than losing a day's earnings to the rain.

Try It Yourself

🔗 Live Prototype → deployed link
📺 Pitch Walkthrough → pitchdeck demo
💻 GitHub → Repository

The difference between those two experiences
is the entire product.

— Kotapothula Siva Raga Adithi
Team Code Alchemists · KL University · DEVTrails 2026

DEV Community: Kotapothula Siva Raga Adithi