GitHub - agdanish/Zynvaro: Zynvaro — AI-Powered Parametric Income Shield for India's Q-Commerce Gig Workers | Guidewire DEVTrails 2026 | Team AeroFyta

When the city stops, your income doesn't.

Zynvaro is a zero-touch parametric insurance platform that automatically detects weather disasters, platform outages, and civic disruptions — and pays gig delivery workers within 5 minutes, with no claims, no forms, no calls.

Built for Guidewire DEVTrails 2026 — Unicorn Chase

Important

What is Parametric Insurance? Unlike traditional insurance where you file a claim and wait, parametric insurance pays out automatically when a measurable trigger crosses a threshold. If IMD rainfall data shows > 64.5mm in your zone — you get paid. Period. No adjuster, no paperwork, no waiting.

📑 Table of Contents

Click to navigate

#	Section	What You'll Learn
🔥	The Crisis	Why 12.7M gig workers need this
🎯	Our Persona	Why Q-Commerce, not food delivery
🎬	Live Scenarios	4 real-world disruption walkthroughs
💰	Premium Model	₹29-₹89/week dynamic pricing
⚡	Parametric Triggers	6 measurable trigger events
🧠	AI/ML Engine	4-layer fraud detection + predictive risk
🛡️	Anti-Spoofing	7-signal authenticity scoring
📱	User Experience	3-screen onboarding, invisible claims
🏗️	Architecture	Full system design + tech choices
📊	Unit Economics	LTV/CAC 9.7x, 62.8% loss ratio
🗺️	Roadmap	Phase 2 & 3 deliverables
🏆	Why We Win	5 unicorn differentiators

     ╔══════════════════════════════════════════════════════════════╗
     ║                                                              ║
     ║   🌧️ Mumbai floods → 3 days lost income → ₹0 recourse       ║
     ║   🏭 Delhi AQI 828 → riders can't breathe → ₹0 recourse     ║
     ║   ☁️ Cloudflare down → Blinkit dies → ₹0 recourse            ║
     ║                                                              ║
     ║   Platform insurance covers accidents.                       ║
     ║   Government covers hospitalization.                         ║
     ║                                                              ║
     ║   ❌ NOBODY covers income loss from external disruptions.    ║
     ║                                                              ║
     ║   ✅ Zynvaro fills this gap.                                 ║
     ║                                                              ║
     ╚══════════════════════════════════════════════════════════════╝

India's 12.7 million gig workers power the digital economy. Yet 80% have zero formal insurance, and no existing product covers income loss from external disruptions.

📉 The Numbers That Define the Crisis

Metric	Data	Source
Gig workers with zero savings	90%	NITI Aayog
Earnings drop during heatwave days	40%	Nature 2024 (Das & Somanathan)
Income loss per 1°C wet-bulb rise	19%	Nature 2024
Q-Commerce GMV (2024)	$6-7B	RedSeer / Bain
Annual heatwave days across India	536	CII / IMD
Delhi AQI > 400 days per winter	30-50	CPCB

Caution

When a disruption hits a Q-Commerce zone, riders don't gradually lose income — they go from full earnings to ₹0 in minutes. The platform algorithm shuts down the zone instantly. No orders = no income = no safety net. Until now.

Why Q-Commerce (Blinkit / Zepto / Instamart) — Not Food Delivery

Tip

Most teams will default to Zomato/Swiggy food delivery riders. We chose Q-Commerce because it is structurally more vulnerable to disruptions — making it the ideal blue-ocean for parametric insurance.

graph LR
    A["🍔 Food Delivery"] -->|"30-45 min SLA"| B["Orders delayed<br>Income reduced"]
    C["⚡ Q-Commerce<br><b>OUR PICK</b>"] -->|"10-15 min SLA"| D["Zone SHUTDOWN<br><b>Income = ₹0</b>"]

    style C fill:#FF6B35,stroke:#fff,color:#fff,stroke-width:2px
    style D fill:#E34F26,stroke:#fff,color:#fff,stroke-width:2px
    style A fill:#555,stroke:#999,color:#fff
    style B fill:#777,stroke:#999,color:#fff

Factor	Food Delivery	Q-Commerce (Our Pick)
Delivery SLA	30-45 min	⚡ 10-15 min
Delivery Radius	Up to 7 km	📍 2-3 km (dark store)
30 min heavy rain impact	Orders delayed	🚫 Entire zone PAUSED
Algorithmic response	Reduced orders	⛔ Instant zone shutdown
Disruption measurability	Moderate	📏 HIGH (tight geofence)

Q-Commerce operates on an ultra-condensed supply chain. When a localized disruption hits — a flooded intersection near a dark store, a sudden AQI spike — the platform's algorithm instantly disables the zone. This "SLA brittleness" makes Q-Commerce the ideal candidate for parametric insurance.

👤 Meet Ravi — Our Target User

Ravi, 27 — Blinkit rider in Koramangala, Bangalore

🛵 Rides a 2-wheeler, works peak + late shift (6 PM - 2 AM)

💰 Nets ~₹18,000-21,000/month after fuel

🏪 Operates from a dark store cluster

💳 Zero savings. Zero insurance.

During Bangalore's October 2024 flooding (157mm in 6 hours), Q-commerce operations were completely halted — Ravi earned ₹0 for 2 days.

What makes Ravi pay ₹49/week?

🪙 Micro-pricing
_{Less than 1 delivery order}

😰 Loss-aversion framing
_{"Your income is at risk"}

🔄 UPI AutoPay
_{Status quo bias}

⚡ Instant value
_{"Covered for tonight"}

✅ Deterministic rules
_{Verify on IMD yourself}

Scenario 1: 🌧️ Monsoon Flooding ^{_{HIGH FREQUENCY}}

Tuesday, 7:45 PM — Ravi starts his evening shift. At 8:30 PM, torrential rain begins — 72mm in 90 minutes. The dark store pauses all orders. Ravi is stuck under a shop awning.

sequenceDiagram
    autonumber
    participant 🌧️ as Weather APIs
    participant ⚙️ as Trigger Engine
    participant 🛡️ as Fraud Pipeline
    participant 💸 as RazorpayX
    participant 📱 as Ravi's WhatsApp

    🌧️->>⚙️: 8:35 PM — Rainfall > 64.5mm detected (IMD threshold)
    ⚙️->>⚙️: Cross-validate with dual-source (OpenWeather + IMD)
    ⚙️->>🛡️: 8:36 PM — Initiate claim for zone riders
    🛡️->>🛡️: GPS ✓ | Device sensors ✓ | Activity ✓ | Score: 92/100
    🛡️->>💸: 8:37 PM — Auto-approved
    💸->>📱: 8:38 PM — ₹300 credited via UPI
    📱->>📱: "Heavy Rain in Koramangala. ₹300 credited. Stay safe! 🌂"

    Note over 🌧️,📱: ⏱️ Total: Disruption → Money in Account = ~5 minutes<br>👆 Zero buttons pressed by Ravi

Scenario 2: ☁️ Platform Outage ^{_{NOVEL — MOST TEAMS WILL MISS THIS}}

Saturday, 1:15 PM — Cloudflare global outage. Blinkit, Zepto, Swiggy APIs return HTTP 503. 200,000+ riders nationwide unable to receive orders.

sequenceDiagram
    autonumber
    participant 🔍 as Synthetic Monitors
    participant 📡 as Downdetector
    participant ⚙️ as Trigger Engine
    participant 💸 as RazorpayX

    🔍->>⚙️: 1:18 PM — HTTP 503 from 5 geo-distributed probes
    📡->>⚙️: 1:19 PM — Downdetector spike confirmed
    ⚙️->>⚙️: 10-min sustained outage → trigger confirmed
    ⚙️->>💸: 1:25 PM — Bulk payout to all active policyholders
    💸->>💸: ₹300 credited to each affected rider

    Note over 🔍,💸: 📌 Platform outage insurance is our UNIQUE differentiator<br>Cloudflare Dec 2025 outage proved this is real

📋 Scenario 3: Severe Air Pollution (Delhi)

November 15, Delhi — AQI hits 485 ("Severe"). GRAP Stage IV activated.

WAQI API + CPCB station both report AQI > 400 for 24 continuous hours
Dual-source validation confirms the event
Riders in affected pincodes receive partial-day payout (₹150) reflecting ~20-30% income impact

🚨 Scenario 4: Coordinated Fraud Attempt (Market Crash)

500 riders in Mumbai organize via Telegram — install GPS-spoofing apps, fake locations into a red-alert weather zone while sitting at home.

Zynvaro's Response: → See Adversarial Defense & Anti-Spoofing Strategy

Spoiler: Our 7-signal authenticity scoring catches them. GPS alone is only 10% of the score.

🔄 End-to-End Application Workflow

flowchart TD
    A["📝 ONBOARDING<br><sub>3 screens, < 60 seconds</sub>"] --> B["🧠 RISK PROFILING<br><sub>AI/ML zone + rider analysis</sub>"]
    B --> C["📋 POLICY CREATION<br><sub>Weekly premium, UPI AutoPay</sub>"]
    C --> D["📡 REAL-TIME MONITORING<br><sub>Weather, AQI, Outage APIs</sub>"]
    D --> E{"⚡ TRIGGER<br>DETECTED?"}
    E -->|"Yes"| F["🛡️ 4-LAYER FRAUD CHECK<br><sub>< 2 minutes</sub>"]
    E -->|"No"| D
    F --> G{"Score<br>≥ 75?"}
    G -->|"Auto-Approve"| H["💸 INSTANT UPI PAYOUT<br><sub>RazorpayX → Rider's UPI</sub>"]
    G -->|"45-74"| I["⏳ Escrow Hold<br><sub>Enhanced verification, 2 hrs</sub>"]
    G -->|"< 45"| J["🔍 Manual Review<br><sub>24-hour queue</sub>"]
    H --> K["📊 ANALYTICS DASHBOARD<br><sub>Loss ratios, trigger feed, fraud flags</sub>"]

    style A fill:#4CAF50,stroke:#fff,color:#fff
    style E fill:#FF9800,stroke:#fff,color:#fff
    style F fill:#9C27B0,stroke:#fff,color:#fff
    style H fill:#2196F3,stroke:#fff,color:#fff
    style K fill:#607D8B,stroke:#fff,color:#fff

Why Weekly — Not Monthly or Annual

Gig workers operate week-to-week. Their platform payouts settle weekly. A ₹200/month premium triggers loss aversion. A ₹49/week premium — deducted on payout day via UPI AutoPay — feels like a platform fee, not an insurance bill.

📋 Coverage Tiers

	🟢 Basic Shield	🔵 Standard Guard	🟣 Pro Armor
Weekly Premium	₹29	₹49	₹89
AQI > 400	✅	✅	✅
Heatwave > 45°C	✅	✅	✅
Heavy Rainfall	❌	✅	✅
Traffic Gridlock	❌	✅	✅
Platform Outage	❌	❌	✅
Civil Disruption	❌	❌	✅
Max Daily Payout	₹300	₹600	₹1,000
Max Weekly Payout	₹600	₹1,200	₹2,000
Target User	Part-time, low-risk city	Full-time, avg-risk city	High-risk weeks (monsoon/pollution)

📐 Dynamic Pricing Formula

Premiums are not static. Our AI recalculates every Monday:

$$ \text{Weekly Premium} = \frac{\text{Expected Loss}}{\text{Target Loss Ratio}} + \text{Risk Loading} $$

Where:

$$ \text{Expected Loss} = \sum_{t} P(\text{trigger}_t \mid \text{zone, week}) \times E(\text{payout}_t \mid \text{tier}) $$

How Ravi's premium changes across seasons:

xychart-beta
    title "Standard Guard Premium — Seasonal Variation (₹)"
    x-axis ["Winter (Dec)", "Pre-Monsoon (May)", "Peak Monsoon (Jul)", "Post-Monsoon (Oct)"]
    y-axis "Premium (₹)" 30 --> 75
    bar [39, 49, 69, 45]

Note

Affordability Guardrail: Premium is hard-capped at 0.8% of estimated weekly net income — ensuring it never becomes unaffordable for the lowest-earning riders.

Resilience Streak Discount: 3 consecutive disruption-free weeks → premium drops by 10%.

Every trigger must be: (a) objectively measurable via public APIs, (b) directly correlated with income loss, and (c) independently verifiable to resist fraud.

#	Trigger	Threshold	Data Sources	Income Loss	Frequency
1	🌧️ Heavy Rainfall	≥ 64.5 mm/24hr (IMD)	OpenWeatherMap + IMD API	40-90%	Mumbai: 15-25 days/monsoon
2	🌊 Extreme Rain / Flooding	≥ 204.5 mm/24hr OR NDMA Red Alert	NDMA SACHET + GDACS	70-100%	Mumbai: ~4 events/year
3	🔥 Severe Heatwave	≥ 45°C for ≥ 2 consecutive days	OpenWeatherMap + IMD Bulletins	20-40%	Delhi: 10-25 days/year
4	🏭 Hazardous AQI	AQI ≥ 401 for 24 continuous hours	WAQI API + CPCB stations	15-30%	Delhi: 30-50 days/winter
5	☁️ Platform Outage	HTTP 503/504 for > 15 min, 3+ probes	Synthetic monitoring + Downdetector	~100%	Cloudflare Dec 2025
6	🚨 Civil Disruption	Section 144 / Curfew ≥ 4 hours	GDELT + NewsAPI + Gazette feeds	60-80%	Multiple events 2023-25

🔎 Deep Dive: Why Platform Outage Insurance is our Blue-Ocean Differentiator

Digital Infrastructure Disruption Index

Beyond weather, we monitor:

Platform outages — synthetic probes + public reporting
Payment rail failures — UPI/IMPS downtime affecting payout settlement
Internet shutdowns — government-mandated, tracked via GDELT + IP traffic analysis

This is credible because:

Blinkit nationwide outage (Feb 2024) — documented
Swiggy Instamart outage (Oct 2024) — documented
Cloudflare Dec 2025 outage → simultaneously killed Blinkit, Zerodha, Groww, and other platforms
Human Rights Watch documents that internet shutdowns directly remove access to app-mediated gig work

No other team will insure digital infrastructure downtime. This is our unique moat.

🏷️ 6.1 — Dynamic Premium Pricing

flowchart LR
    A["📊 Feature Space<br><sub>Geospatial, Temporal,<br>Environmental, Behavioral,<br>Integrity signals</sub>"] --> B["📈 Tweedie GLM<br><sub>Actuarial base rate</sub>"]
    A --> C["🌲 XGBoost<br><sub>Non-linear risk factors</sub>"]
    B --> D["🔀 Ensemble"]
    C --> D
    D --> E["💡 SHAP Explainer<br><sub>Waterfall charts:<br>'Why is your premium ₹X?'</sub>"]
    E --> F["💰 Personalized<br>Weekly Premium<br><sub>₹29 — ₹89</sub>"]

    style D fill:#FF9800,stroke:#fff,color:#fff
    style F fill:#4CAF50,stroke:#fff,color:#fff

📋 Complete Feature Space (All measurable in hackathon)

Category	Features
Geospatial	Rider's operational pincode, H3 cell (Res 7-8), elevation proxy, distance to coast/floodplain
Temporal	Week-of-year, monsoon flag, festival indicator, day-length
Environmental	7-day weather forecast, trailing AQI averages, seasonal pollution patterns
Rider Behavior	Declared shift window, avg active hours, claim history, "online but stationary" ratio
Integrity	Device attestation status, GPS accuracy radius, spoof-risk score

🛡️ 6.2 — Intelligent Fraud Detection (4-Layer Pipeline)

flowchart TD
    subgraph "⚡ LAYER 1 — Real-Time Telemetry"
        L1["🔍 Isolation Forest<br><sub>GPS × Wi-Fi × Cell Tower × IP</sub><br><sub>Catches spoofing in sub-seconds</sub>"]
    end

    subgraph "🧠 LAYER 2 — Behavioral Anomaly"
        L2["📊 Autoencoder + LSTM<br><sub>Accelerometer/gyroscope time-series</sub><br><sub>High reconstruction error = anomaly</sub>"]
    end

    subgraph "🕸️ LAYER 3 — Network Collusion"
        L3["🔗 Graph Neural Networks<br><sub>Louvain community detection</sub><br><sub>40 riders on same IP = syndicate</sub>"]
    end

    subgraph "✅ LAYER 4 — Cross-Reference"
        L4["🌲 XGBoost Classifier<br><sub>IMD + CPCB + GDACS + TomTom</sub><br><sub>Multi-source consensus required</sub>"]
    end

    L1 --> L2 --> L3 --> L4
    L4 --> RESULT{"Authenticity<br>Score"}

    style L1 fill:#2196F3,stroke:#fff,color:#fff
    style L2 fill:#9C27B0,stroke:#fff,color:#fff
    style L3 fill:#FF5722,stroke:#fff,color:#fff
    style L4 fill:#4CAF50,stroke:#fff,color:#fff
    style RESULT fill:#FF9800,stroke:#fff,color:#fff

🔮 6.3 — Predictive Risk Engine (Competitive Moat)

Architecture: LSTM (48-hour lookback, 7 features, 2 layers of 64+32 units) + XGBoost ensemble

Inputs: Multi-model weather forecasts, seasonal climatology, historical trigger counts, AQI patterns, NDMA alerts

Two strategic outputs:

📈 Underwriting discipline — premiums adjust BEFORE monsoon spikes, protecting the loss ratio
🔔 Preventive intelligence alerts — WhatsApp: "Heatwave expected in your zone tomorrow. Consider the evening shift."

Important

🦄 Unicorn Feature — Preventive Payouts

If severe heat is predicted 48 hours ahead, auto-disburse ₹50 to the rider's wallet earmarked for ORS (Oral Rehydration Salts) and water.

By investing ₹50 proactively, we prevent the much larger ₹600 claim — transforming insurance from reactive to proactive.

This is insurance reimagined. Not just paying after disaster — preventing the loss entirely.

Warning

Market Crash Scenario: 500 delivery workers coordinate via Telegram, install GPS-spoofing apps, fake locations into a red-alert weather zone while resting at home — attempting to drain the liquidity pool.

Our response: GPS is only 10% of our authenticity score. We have 6 other signals they can't fake.

7-Signal Authenticity Scoring Matrix

Signal	What It Detects	Why Spoofers Can't Fake It	Weight
📍 GPS	Basic location	Easily spoofed — baseline only	10%
📶 Wi-Fi BSSID	Nearby Wi-Fi networks	Spoofing GPS doesn't change which routers your phone sees	20%
📡 Cell Tower ID	Tower connections	Hardware-level; GPS apps don't alter cellular connections	20%
🌐 IP Geolocation	ISP routing location	GPS says "Andheri" but IP resolves to "Thane" = caught	10%
📳 Accelerometer	Physical motion	Stranded rider shows micro-movements; home rider = flatline	20%
🌡️ Barometric Pressure	Altitude + weather	Real rainstorm = pressure drops; dry apartment = normal	10%
📊 Network Latency	Connection quality	Storm zones show degraded signal; home Wi-Fi = pristine	10%

    Composite Authenticity Score: 0 ──────────────────────────── 100

    ❌ AUTO-REJECT     🔍 MANUAL REVIEW     ⏳ ESCROW      ✅ AUTO-APPROVE
    ├─────────────────┤├──────────────────┤├─────────────┤├───────────────┤
    0                 24                  44             75              100

🕸️ How We Catch the 500-Rider Syndicate

Graph-Based Collusion Detection

flowchart LR
    subgraph "Rider Network Graph"
        R1["Rider A"] --- R2["Rider B"]
        R2 --- R3["Rider C"]
        R1 --- R3
        R3 --- R4["Rider D"]
        R4 --- R5["Rider E"]
    end

    subgraph "Suspicious Edges"
        E1["Same IP Subnet"]
        E2["Shared Device ID"]
        E3["Same Bank Account"]
        E4["Co-registration < 48hrs"]
        E5["Claims within 60 sec"]
    end

    subgraph "Detection"
        D1["🔍 Louvain Algorithm"]
        D2["📊 Label Propagation"]
        D3["⚠️ Anomaly Score"]
    end

    R1 -.-> E1
    R2 -.-> E2
    R3 -.-> E3
    E1 --> D1
    E2 --> D2
    E3 --> D3

Signal	Normal Pattern	🚨 Fraud Ring Pattern	Detection
Claim timing	Spread across hours	500 claims in 60-second window	Burst detection
IP addresses	Diverse ISPs	Same subnet cluster	Entropy analysis
Device fingerprints	Unique per rider	Shared IMEI/cloned apps	Hash collision
Referral chains	Organic, varied	Linear chain from 1 source	Graph depth analysis
Claim-to-registration	Claims after weeks	Claims within days of signup	Velocity scoring

The Telegram Tell: 500 riders claiming from the "same zone" within 60 seconds — but cell tower data shows they're in 200+ different locations. A genuine weather event causes claims to trickle over 30-60 minutes. Instantaneous mass claims are a statistical impossibility in organic disruption.

⚖️ UX Balance — Protecting Honest Workers

Graduated Response Protocol

Score	Action	Rider Sees
75-100 ✅	Auto-approve, instant payout	"₹300 credited. Stay safe! 🌂"
45-74 ⏳	Escrow hold (2 hrs), enhanced verification	"Processing. Confirmed within 2 hours." + optional selfie/screenshot
25-44 🔍	Manual review (24 hrs)	"We need more time to verify. You'll hear within 24 hours."
0-24 ❌	Soft block + investigation	"Couldn't verify disruption. Tap to request review."

Critical Principles:

Never punish network drops — use last-known-good location + zone-level disruption confirmation
Benefit-of-the-doubt — clean-history riders get payouts with soft flag, not hold
Appeal mechanism — one-tap "Request Review" + 10% bonus for wrongful rejections
Syndicate isolation — punish the network, not the neighborhood

Onboarding: 3 Screens, < 60 Seconds

flowchart LR
    S1["📍 Screen 1<br><b>Protect This Week</b><br><sub>Auto-detect city/zone<br>OTP login<br>Language select</sub>"] --> S2["🛡️ Screen 2<br><b>Pick Your Shield</b><br><sub>Basic / Standard / Pro<br>No insurance jargon<br>Social proof</sub>"] --> S3["💸 Screen 3<br><b>Payout Setup</b><br><sub>UPI ID / bank account<br>AutoPay mandate<br>'You're Protected' ✅</sub>"]

    style S1 fill:#4CAF50,stroke:#fff,color:#fff
    style S2 fill:#2196F3,stroke:#fff,color:#fff
    style S3 fill:#9C27B0,stroke:#fff,color:#fff

Claims Flow — Completely Invisible to Rider

Target: Disruption → Money-in-account: < 30 minutes for auto-approved claims.

stateDiagram-v2
    [*] --> DisruptionOccurs: 🌧️ Weather / ☁️ Outage / 🚨 Curfew
    DisruptionOccurs --> TriggerDetected: Dual-source API polling (every 5 min)
    TriggerDetected --> PolicyQuery: Which riders are covered in this H3 zone?
    PolicyQuery --> FraudPipeline: 4-layer authenticity scoring (< 2 min)
    FraudPipeline --> AutoApprove: Score ≥ 75
    FraudPipeline --> EscrowHold: Score 45-74
    FraudPipeline --> ManualReview: Score < 45
    AutoApprove --> UPIPayout: 💸 RazorpayX instant transfer
    UPIPayout --> WhatsAppNotify: 📱 "₹300 credited. Stay safe!"
    EscrowHold --> EnhancedVerification: 2-hour review window
    ManualReview --> AppealQueue: 24-hour review + appeal option
    WhatsAppNotify --> [*]

Platform Architecture

Audience	Platform	Why
🛵 Riders	PWA + WhatsApp Bot	Zero-install, works on low-end Android, 500M+ Indian WhatsApp users
📊 Insurers	Web Dashboard	Loss ratios, fraud analytics, trigger feed, premium volume

System Architecture

flowchart TB
    subgraph FRONTEND["🖥️ FRONTEND LAYER"]
        F1["⚛️ React PWA<br><sub>Rider App + Tailwind</sub>"]
        F2["📊 React Dashboard<br><sub>Admin + Deck.gl + H3</sub>"]
        F3["💬 WhatsApp<br><sub>Cloud API</sub>"]
    end

    subgraph API["⚡ API GATEWAY — FastAPI"]
        A1["🔐 Auth"]
        A2["📋 Policy Mgmt"]
        A3["💰 Premium Engine"]
        A4["📝 Claims"]
        A5["💸 Payouts"]
    end

    subgraph ML["🧠 ML SERVICES — Python"]
        M1["📈 XGBoost + LightGBM<br><sub>Pricing engine</sub>"]
        M2["🔮 LSTM<br><sub>Predictive risk</sub>"]
        M3["💡 SHAP<br><sub>Explainability</sub>"]
    end

    subgraph TRIGGER["📡 TRIGGER ENGINE"]
        T1["🔄 API Poller"]
        T2["📨 Redis Streams"]
    end

    subgraph FRAUD["🛡️ FRAUD PIPELINE"]
        FR1["🔍 Isolation Forest"]
        FR2["📊 Autoencoder"]
        FR3["🕸️ GNN (Louvain)"]
        FR4["🌲 XGBoost"]
    end

    subgraph DATA["💾 DATA LAYER"]
        D1["🐘 PostgreSQL<br><sub>+ PostGIS + h3-pg</sub>"]
        D2["⚡ Redis<br><sub>Real-time cache</sub>"]
    end

    subgraph EXTERNAL["🌐 EXTERNAL APIs"]
        E1["🌤️ OpenWeatherMap"]
        E2["🏭 WAQI / CPCB"]
        E3["🌊 NDMA SACHET"]
        E4["📰 GDELT / NewsAPI"]
        E5["💸 RazorpayX"]
        E6["💬 WhatsApp API"]
    end

    FRONTEND --> API
    API --> ML
    API --> TRIGGER
    API --> FRAUD
    ML --> DATA
    TRIGGER --> DATA
    FRAUD --> DATA
    TRIGGER --> EXTERNAL
    API --> EXTERNAL

Tech Stack at a Glance

Layer	Technology	Why
Frontend		Fast, mobile-first, offline-capable
Admin		H3 hexagonal risk maps in WebGL
Backend		Async, high-performance, ML-native
ML/AI		Gradient boosting + deep learning
Database		Geospatial + ACID for financial records
Cache		Real-time zone scores + event streaming
Payments		Instant UPI payouts with idempotency
Messaging		500M+ Indian users
Monitoring		Trigger latency + system health
Deploy		Zero-infrastructure management

📡 External APIs — All Free/Freemium Tier

API	Free Tier	Use Case
OpenWeatherMap 3.0	1,000 calls/day	Precipitation, temperature, severe alerts
WeatherAPI.com	1M calls/month	Backup weather + built-in AQI
WAQI (aqicn.org)	1,000 req/sec	Real-time AQI from CPCB stations
OpenAQ v3	Unlimited	Historical air quality for model training
NDMA SACHET	Free (RSS/JSON)	Cyclone, flood, earthquake CAP alerts
GDACS	Free (6-min updates)	International disaster alerts, flood severity
GDELT Project	Free (BigQuery)	Civil disruption, protest, curfew detection
NewsAPI	100 req/day	Strike/curfew news corroboration
RazorpayX	Test mode (free)	UPI payout simulation
WhatsApp Cloud API	1,000 conv/month	Notifications + onboarding

Weekly P&L — 10,000 Riders

pie title Premium Collection by Tier (Weekly)
    "🟢 Basic Shield (5,000 × ₹29)" : 145000
    "🔵 Standard Guard (3,500 × ₹49)" : 171500
    "🟣 Pro Armor (1,500 × ₹89)" : 133500

	Collection	Claims	Loss Ratio
🟢 Basic (50%)	₹1,45,000	₹75,000	51.7%
🔵 Standard (35%)	₹1,71,500	₹1,57,500	91.8%
🟣 Pro (15%)	₹1,33,500	₹1,50,000	112.4%
📊 Portfolio	₹4,50,000	₹2,82,500	62.8% ✅

Target Loss Ratio: 60-65% — our 62.8% sits right in the sweet spot.

🦄 Key Metrics

Metric	Value	Benchmark
Customer Acquisition Cost	₹45	WhatsApp referrals + dark-store partnerships
Average Weekly Premium	₹45	Blended across tiers
Average Retention	26 weeks	Conservative (65% annual gig attrition)
Lifetime Value (LTV)	₹435	$45 × 26 × (1 - 0.628)$
LTV / CAC Ratio	9.7x	Benchmark: > 3x is venture-scale
Break-even Subscribers	~5,000	Achievable in first 3 months

🌧️ Monsoon Stress Test

During peak monsoon, trigger probability rises to 25-35%. The dynamic pricing engine responds:

Premium auto-adjusts — Standard: ₹49 → ₹69 for high-risk zones
Episode caps — max 4 disruption windows/week
Smart-reserve pool — winter surplus funds monsoon payouts

Phase 2: Automation & Protection _{Mar 21 - Apr 4}

Registration and onboarding flow (3-screen PWA)
Insurance policy management (CRUD + weekly renewal engine)
Dynamic premium calculation (XGBoost pricing model)
5 automated parametric triggers with real API integrations
Basic claims management with fraud Layer 1 (GPS validation)
RazorpayX test-mode payout integration
2-minute demo video

Phase 3: Scale & Optimize _{Apr 5 - 17}

Advanced fraud detection (all 4 layers — Autoencoder, LSTM, GNN, XGBoost)
Instant payout system (simulated end-to-end)
Predictive risk engine with next-week forecasting
Worker dashboard (earnings protected, active coverage, trust score)
Admin dashboard (loss ratios, trigger feed, fraud flags, analytics)
Evidence Bundle for every claim (multi-source proof card)
5-minute demo video + final pitch deck

gantt
    title Zynvaro Development Timeline
    dateFormat YYYY-MM-DD
    axisFormat %b %d

    section Phase 1 — Seed
    Ideation & Research          :done, p1a, 2026-03-04, 2026-03-15
    README & Repo Setup          :done, p1b, 2026-03-15, 2026-03-20
    Phase 1 Submission           :milestone, m1, 2026-03-20, 0d

    section Phase 2 — Scale
    Registration & Onboarding    :p2a, 2026-03-21, 4d
    Policy Management CRUD       :p2b, 2026-03-25, 3d
    Dynamic Premium Engine       :p2c, 2026-03-25, 5d
    Parametric Trigger System    :p2d, 2026-03-28, 4d
    Claims + Fraud Layer 1       :p2e, 2026-03-30, 3d
    RazorpayX Integration        :p2f, 2026-04-01, 2d
    Phase 2 Submission           :milestone, m2, 2026-04-04, 0d

    section Phase 3 — Soar
    Advanced Fraud (4 Layers)    :p3a, 2026-04-05, 5d
    Predictive Risk Engine       :p3b, 2026-04-07, 4d
    Dashboards (Worker + Admin)  :p3c, 2026-04-10, 4d
    Evidence Bundles + Payouts   :p3d, 2026-04-12, 3d
    Demo Video + Pitch Deck      :p3e, 2026-04-14, 3d
    Phase 3 Submission           :milestone, m3, 2026-04-17, 0d

What Makes This a ⭐⭐⭐⭐⭐, Not a ⭐⭐⭐

Dimension	⭐⭐⭐ Meets Brief	⭐⭐⭐⭐⭐ Zynvaro
Architecture	Single weather trigger, cron polling, monolithic backend	Multi-source event-driven pipeline, H3 spatial grid, 4-layer fraud ML, streaming
UX	Download app, fill forms, press "Claim Now"	Zero-install PWA + WhatsApp, 3-screen onboard, zero-touch claims, money in 5 min
Logic	"It rained, so pay everyone"	IMD/CPCB official thresholds, dynamic pricing with SHAP explainability, 62% loss ratio, LTV/CAC 9.7x

🦄 Our 5 Unicorn Differentiators

☁️
Platform Outage Insurance

_{We insure digital infrastructure downtime. Cloudflare Dec 2025 outage proves it's real.}

🔮
Preventive Payouts

_{AI predicts heatwave 48hrs ahead, sends ₹50 for ORS. Prevents the ₹600 claim.}

📋
Evidence Bundles

_{Every payout shows a "Proof Card" with 2+ weather sources, AQI, zone match, logs.}

🗺️
H3 Hexagonal Grid

_{Uber's H3 + Deck.gl WebGL for hyper-local zone risk visualization.}

🛡️
Anti-Spoofing

_{7-signal authenticity scoring + graph syndicate detection. Market Crash proof.}

👥 Team AeroFyta

	Name	Role
👨‍💻	Danish A G	Team Lead
👨‍💻	Sanjay N	Developer
👨‍💻	Athishaya K	Developer
👨‍💻	Vishal C B	Developer
👨‍💻	Hariharan C V	Developer

Zynvaro — Because every delivery matters. Every rider deserves a safety net.

_{Built with conviction by Team AeroFyta for Guidewire DEVTrails 2026 — Unicorn Chase}

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
assets/headers		assets/headers
docs		docs
logo		logo
.gitignore		.gitignore
README.md		README.md
index.html		index.html

Folders and files

Latest commit

History

Repository files navigation

When the city stops, your income doesn't.

📑 Table of Contents

📉 The Numbers That Define the Crisis

Why Q-Commerce (Blinkit / Zepto / Instamart) — Not Food Delivery

👤 Meet Ravi — Our Target User

Scenario 1: 🌧️ Monsoon Flooding HIGH FREQUENCY

Scenario 2: ☁️ Platform Outage NOVEL — MOST TEAMS WILL MISS THIS

🔄 End-to-End Application Workflow

Why Weekly — Not Monthly or Annual

📋 Coverage Tiers

📐 Dynamic Pricing Formula

Digital Infrastructure Disruption Index

🏷️ 6.1 — Dynamic Premium Pricing

🛡️ 6.2 — Intelligent Fraud Detection (4-Layer Pipeline)

🔮 6.3 — Predictive Risk Engine (Competitive Moat)

7-Signal Authenticity Scoring Matrix

Graph-Based Collusion Detection

Graduated Response Protocol

Onboarding: 3 Screens, < 60 Seconds

Claims Flow — Completely Invisible to Rider

Platform Architecture

System Architecture

Tech Stack at a Glance

Weekly P&L — 10,000 Riders

🦄 Key Metrics

Phase 2: Automation & Protection Mar 21 - Apr 4

Phase 3: Scale & Optimize Apr 5 - 17

What Makes This a ⭐⭐⭐⭐⭐, Not a ⭐⭐⭐

🦄 Our 5 Unicorn Differentiators

👥 Team AeroFyta

Zynvaro — Because every delivery matters. Every rider deserves a safety net.

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Scenario 1: 🌧️ Monsoon Flooding ^{_{HIGH FREQUENCY}}

Scenario 2: ☁️ Platform Outage ^{_{NOVEL — MOST TEAMS WILL MISS THIS}}

Phase 2: Automation & Protection _{Mar 21 - Apr 4}

Phase 3: Scale & Optimize _{Apr 5 - 17}

Packages