Skip to content

dhanushkrishnac7/GuideWire

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

36 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

GigWire — AI-Powered Parametric Income Insurance for India's Gig Economy

Guidewire DEVTrails 2026 | Unicorn Chase | Phase 2 Submission _ Protecting the last-mile heroes of India's digital economy _


Table of Contents

  1. Problem Statement & Market Context
  2. Our Solution — GigWire
  3. Persona Focus
  4. Insurance Domain — Coverage, Exclusions & Regulatory Framework
  5. Actuarial Model & Financial Sustainability
  6. System Architecture
  7. Core Modules & Algorithms
  8. Adversarial Defense & Anti-Spoofing Strategy
  9. Basis Risk Mitigation — Ward-Level Data Architecture
  10. Key Innovation: Privacy-First API Integration
  11. Predictive Alert & Reward System
  12. Application Screens
  13. Admin Dashboard — Insurer Web Portal
  14. Simulation Screen — Live Verification Demo
  15. Tech Stack
  16. Platform Strategy — Mobile App + Embedded SDK
  17. Weekly Pricing Model
  18. Data Flow Diagrams
  19. Research References

1. Problem Statement & Market Context

The Protection Gap

India's platform-based delivery partners — working for Zomato, Swiggy, Zepto, Amazon, Dunzo, and Blinkit — are among the fastest-growing and most financially precarious segments of the Indian workforce. According to NITI Aayog, the gig workforce grew by approximately 55% over a recent four-year period, reaching roughly 12 million platform workers by 2024–2025, accounting for over 2% of the entire Indian workforce. Projections estimate this will reach 23.5 million by 2030. Delivery partners across food delivery and quick commerce platforms alone process over 10 million daily orders.

Despite this scale, these workers operate entirely outside traditional social security systems as independent contractors — with zero income protection against external disruptions.

The Financial Reality (2026 Data)

Platform Segment Gross Monthly Income Net After Expenses Payout Cycle
Food Delivery (Zomato / Swiggy) ₹26,500–₹27,700 ~₹21,000 Weekly
Quick Commerce (Zepto / Blinkit) ₹25,000–₹35,000 ~₹18,000–₹26,000 Weekly
E-commerce (Amazon / Flipkart) ₹20,000–₹30,000 ~₹15,000–₹22,000 Varies

Approximately 20–32% of gross income is consumed by fuel, vehicle maintenance, mobile data, and asset insurance — leaving narrow net margins that are acutely vulnerable to disruption. External events such as extreme weather, severe air pollution, and localized unrest cause a 20–30% reduction in monthly earnings for delivery partners.

Critically, missing a peak earning window — typically between 7 PM and 10 PM — does not merely cost that window's wages. Platform algorithms prioritize consistent, active workers, so a missed peak window also causes a worker to miss daily and weekly incentive targets and negatively impacts future priority order allocation. The income loss is non-linear.

Why Existing Parametric Products Have Failed

India has seen several parametric insurance pilots — SEWA/ICICI Lombard for extreme heat, Digit Insurance for AQI-based construction worker coverage, Bajaj Allianz for multi-peril climate triggers (2025). These programs have stalled in a pre-scaling loop due to:

  • Basis risk — coarse city-level data failing to reflect on-ground conditions (the Tata AIG/Nagaland program failed to pay a single claim during actual floods because thresholds relied on satellite data instead of ward-level readings)
  • Affordability — premium costs negating already narrow profit margins
  • No social disruption coverage — weather APIs cannot detect strikes, curfews, or market closures

GigWire is designed to directly address each of these failures.

External disruptions covered by GigWire:

Disruption Type Examples Worker Impact
Environmental Heavy rain (>50mm/hr), Floods, Extreme heat (>45°C), AQI >300 Cannot work outdoors / Deliveries halted
Social Unplanned curfews, Local strikes, Sudden market/zone closures Unable to access pickup/drop locations

2. Our Solution — GigWire

GigWire is a lightweight, AI-powered parametric income insurance platform that:

  • Detects disruption events automatically using 5 automated triggers across weather, AQI, traffic, and news APIs — at ward level, not city level
  • Verifies every event through a mandatory two-stage agentic AI pipeline before any payout is triggered
  • Guards against GPS spoofing, fake location claims, and coordinated fraud using a multi-layer adversarial defense system
  • Calculates payout based on the worker's actual lost hours and declared income
  • Pays instantly via simulated UPI/Razorpay within minutes of event confirmation
  • Rewards workers who act on predictive disruption alerts with premium discounts

3. Persona Focus

Primary Persona: Food Delivery Partners — Zomato & Swiggy

  • Work 6–14 hours/day on two-wheelers, earn ₹600–₹1,200/day (net ~₹21,000/month)
  • Own low-end Android phones (2GB RAM, Android 8+), operate in 2G/3G areas
  • Skeptical of insurance — need rupee-value ROI shown plainly, not policy jargon
  • Speak Hindi or regional languages primarily
  • Financially exposed during peak hours (7–10 PM) when algorithmic incentives compound loss

4. Insurance Domain — Coverage, Exclusions & Regulatory Framework

GigWire is a parametric micro-insurance product designed in accordance with IRDAI regulations for micro-segment insurance products targeting unorganised workers.

What Is Covered

Coverage is strictly limited to verifiable income loss caused by external disruptions that prevent a delivery worker from working. A payout is triggered only when all three conditions are met:

  1. A qualifying disruption event is detected and verified at ward level in the worker's active zone
  2. The worker was confirmed as active (online and in-zone) during the disruption window
  3. The claim passes all fraud and anti-spoofing checks
Covered Event Parametric Trigger Threshold
Heavy Rain / Flood Rainfall > 50mm/hr OR IMD Red Alert active in zone
Extreme Heat Temperature > 45°C AND Heat Index > 52°C
Severe Air Pollution AQI > 300 (Hazardous, CPCB classification)
Unplanned Curfew NLP-confirmed government curfew order for zone
Local Strike / Bandh NLP-confirmed + corroborated across ≥2 independent news sources
Market / Zone Closure Traffic velocity drop below threshold + zone-specific NLP signal

Standard Exclusions — Explicitly Enforced in Policy UI

Every GigWire policy displays these exclusions on the Policy Details screen. The worker must acknowledge them during onboarding before coverage activates.

Exclusion Category Specific Exclusions
Health & Medical Illness, hospitalisation, COVID-19, personal injury during work
Life & Disability Death benefit, permanent disability, critical illness
Accident & Liability Road accidents, third-party liability, collision damage
Vehicle & Equipment Bike repair, fuel costs, tyre replacement, mobile damage
Intentional Acts Self-caused disruptions, deliberate avoidance of work
Pre-existing Conditions Known scheduled strikes, pre-announced events, planned maintenance
Income Beyond Cap Any loss exceeding 5× the weekly premium paid
Non-active Workers Workers who were offline or outside the disruption zone

Policy Principle: GigWire insures _ income lost during an uncontrollable external disruption _ only. It does not insure the worker's health, vehicle, or any recovery cost. The claim engine is structurally incapable of generating payouts for excluded categories.

Regulatory Framework

IRDAI Compliance: GigWire is designed under the IRDAI Insurance Products Regulations, which require all insurance products to be simple to understand, transparent in exclusions, and actuarially justified. As a micro-insurance product for unorganised workers with low premium rates, GigWire adheres to IRDAI's micro-insurance regulations.

IRDAI Sandbox Framework (2019): GigWire's operational pathway is the IRDAI Regulatory Sandbox, which allows InsurTech platforms to test automated parametric models on live populations without completing full product approval cycles. This framework was explicitly created to enable hyper-local, worker-focused parametric products.

Sabka Bima Sabki Raksha Act (2025): The landmark legislation passed in late 2025 amending the Insurance Act of 1938 explicitly identifies gig workers and informal sector participants as underinsured segments and promotes parametric and micro-insurance as a national priority. The act also authorised a dedicated Policyholders' Education and Protection Fund. GigWire operates at the exact intersection this legislation was designed to address.

Weekly Pricing Mandate

All premiums are structured on a weekly basis only — aligned with the payout cycle of a gig worker. No monthly, quarterly, or annual premium options exist in the system.


5. Actuarial Model & Financial Sustainability

Burning Cost Rate (BCR)

For GigWire to operate sustainably without reserve drainage, insurance operations are modelled on continuous monitoring of the Burning Cost Rate:

BCR = Total Claims Paid / Total Premiums Collected

Target BCR for a sustainable parametric model: 0.55 to 0.70 — meaning approximately 65 paise of every rupee collected goes toward payouts, with the remainder covering operations and profit margins.

If the aggregate loss ratio across any specific zone pool exceeds 85%, actuarial guidelines mandate the suspension of new enrollments for that pool to prevent reserve drainage.

Actuarial Premium Formula

The academically grounded premium formula from parametric insurance literature:

Premium = (Trigger Probability) × (Average Income Lost Per Day)

This is the base actuarial calculation. GigWire extends this with zone-specific, worker-specific, and seasonal modifiers (see Section 7.2) to produce a hyper-local weekly premium that reflects actual on-ground risk rather than city-level averages.

Stress Testing

To withstand prolonged climate events, the premium model is stress-tested against at least one severe scenario — specifically a continuous 14-day monsoon period — ensuring adequate reserve capital to cover peak liability across all active zone pools simultaneously.

Reinsurance Consideration

To protect against extreme volatility from high-frequency events, the GigWire model is structured to support reinsurance layering in production — transferring excess liability to reinsurers above defined thresholds. This eliminates binary cliff effects where minor fluctuations flip a payout between 100% and 0%, ensuring capital efficiency and liquidity precisely when workers need it most.


6. System Architecture

6.1 High-Level Architecture

graph TB
    subgraph External_Data["External Data Sources"]
        WA[OpenWeatherMap / Meteomatics]
        AQI[CPCB AQI API]
        TA[Mapbox Traffic API]
        NF[NewsAPI / Local RSS Feeds]
        GOV[NDMA / State Alerts]
    end

    subgraph Trigger_Engine["Trigger Engine — 5 Automated Triggers"]
        T1[Trigger 1: Heavy Rain / Flood]
        T2[Trigger 2: Extreme Heat]
        T3[Trigger 3: Hazardous AQI]
        T4[Trigger 4: Traffic / Zone Closure]
        T5[Trigger 5: Strike / Curfew NLP]
    end

    subgraph Verification["Two-Step Agentic Verification"]
        A1[Agent 1: Structural Validator]
        A2[Agent 2: Cross-Source Corroborator]
        CE[Confidence Engine: CONFIRM / WATCH / DISMISS]
    end

    subgraph AntiSpoof["Adversarial Defense Layer"]
        GPS[GPS Spoof Detector]
        RG[Rule Guards — Sync Checks]
        FD[Isolation Forest Fraud Scorer]
    end

    subgraph Core["GigWire Core Platform"]
        PM[Premium Engine: XGBoost + GScore]
        CL[Claims Engine]
        PY[Payout Processor: Razorpay + NPCI UPI]
        NE[Notification Engine: FCM]
    end

    subgraph App["GigWire App — React Native + Expo"]
        OB[Onboarding]
        DB[Dashboard + What-If Panel]
        PL[Policy Management]
        CR[Claims & Payouts]
        IN[Insights & Savings]
        SIM[Simulation Screen]
        ADM[Admin Dashboard]
    end

    subgraph SDK["Partner SDK"]
        ZQ[Zone Query API — Privacy-First]
        WH[Webhook Event Notifier]
    end

    WA & AQI & TA & NF & GOV --> T1 & T2 & T3 & T4 & T5
    T1 & T2 & T3 & T4 & T5 --> A1 --> A2 --> CE
    CE -->|CONFIRMED| CL
    CE -->|CONFIRMED| NE
    CL --> RG --> GPS --> FD
    FD -->|Score < 0.3| PY
    FD -->|Score 0.3–0.6| CL
    PM --> App
    PY --> App
    NE --> App
    ZQ --> CL
    WH --> SDK
Loading

6.2 Phase 2 Changes from Phase 1

Component Phase 1 Phase 2
Trigger count Conceptual 5 fully implemented automated triggers
Verification Designed Live two-step agentic pipeline with confidence tiers
Fraud detection Isolation Forest spec Isolation Forest + GPS spoof detector fully implemented
Anti-spoofing Not present Full adversarial defense layer including RF fingerprinting
Data granularity City-level Ward-level / pincode-level (basis risk mitigation)
Actuarial model Basic formula BCR tracking + stress-tested premium model
Simulation Not present Interactive simulation screen with live algorithm logs
Home screen Static Dynamically updates when any trigger fires
Policy screen Basic Explicit exclusions list, IRDAI-aligned, weekly pricing enforced
Regulatory Not addressed IRDAI Sandbox + Sabka Bima Sabki Raksha Act alignment

7. Core Modules & Algorithms

7.1 GScore — Worker Risk Profiling

The GScore is a composite worker performance metric used for premium adjustment and fraud credibility scoring. It ensures experienced, resilient workers pay lower premiums and that claims from low-GScore workers receive closer scrutiny.

Reference: Gohil, J. & Jha, A. (2024). _ Addressing Policy Gaps for Gig Workers in India. _ IJFMR, Vol. 6, Issue 6. E-ISSN: 2582-2160. Section 4.2.

Formula (weights are fixed per the reference — do not modify):

GScore = (0.15 × Time) + (0.20 × Orders) + (0.15 × BeforeTime)
       + (0.15 × BadWeather) + (0.20 × Distance) + (0.15 × TimeOfDay)

All parameters normalised to [0, 1].

Parameter Weight Description
Time 0.15 Delivery time efficiency
Orders 0.20 Daily order volume
BeforeTime 0.15 On-time delivery rate
BadWeather 0.15 Orders completed in adverse conditions
Distance 0.20 Average distance per delivery
TimeOfDay 0.15 Deliveries during challenging hours (late night / early morning)
GScore Range Worker Tier Premium Modifier
0.75–1.00 Elite −15%
0.50–0.74 Reliable −5%
0.30–0.49 Standard Base
0.00–0.29 At-Risk +10% loading

7.2 Dynamic Premium Calculation

Algorithm: XGBoost (Gradient Boosting) trained on 3-year historical disruption data per pincode, with monthly online learning updates. All data inputs are resolved at ward / pincode level — never city-level averages.

GigWire premium formula (extended from actuarial base):

Weekly Premium = Base_Premium × Zone_Risk_Multiplier × Season_Factor × GScore_Modifier
Zone_Risk_Multiplier = normalize(
    0.40 × weather_risk_score(ward, week)
  + 0.25 × flood_zone_index(pincode)
  + 0.20 × pollution_index_7day_avg(ward)
  + 0.15 × strike_history_score(zone, month)
)

Season_Factor = historical_disruption_freq(zone, month) / annual_avg
GScore_Modifier = 1 - ((GScore - 0.5) × 0.3)   // range: 0.85–1.15

Worked example — Chennai, December, reliable worker:

Base: ₹35
Zone_Risk: 1.45  (cyclone-prone coastal ward)
Season:    1.60  (December = 3× annual average)
GScore:    0.88  (Reliable tier)
Premium  = ₹35 × 1.45 × 1.60 × 0.88 ≈ ₹71/week

Live data inputs (Phase 2 integrations):

  • OpenWeatherMap + Meteomatics → ward-level weather risk
  • CPCB AQI API → ward-level pollution index
  • Mapbox → flood-prone zone tagging, traffic density
  • Historical pincode data → waterlogging frequency, strike history

7.3 Disruption Detection Pipeline & Automated Triggers

Five automated triggers implemented in Phase 2 using triggerEngine.ts and triggerConfig.json. All thresholds are evaluated at ward / pincode level.

flowchart LR
    subgraph T1["Trigger 1 — Rain/Flood"]
        R1[OpenWeatherMap API] --> R2{rainfall > 50mm/hr<br/>OR IMD Red Alert?}
        R2 -->|YES| R3[RAIN_FLOOD event]
    end
    subgraph T2["Trigger 2 — Heat"]
        H1[Meteomatics API] --> H2{temp > 45°C<br/>heat_index > 52°C?}
        H2 -->|YES| H3[EXTREME_HEAT event]
    end
    subgraph T3["Trigger 3 — AQI"]
        A1[CPCB AQI API] --> A2{AQI > 300<br/>Hazardous?}
        A2 -->|YES| A3[HAZARDOUS_AQI event]
    end
    subgraph T4["Trigger 4 — Traffic/Closure"]
        M1[Mapbox API] --> M2{zone avg speed<br/>< critical threshold?}
        M2 -->|YES| M3[ZONE_CLOSURE event]
    end
    subgraph T5["Trigger 5 — Strike/Curfew"]
        N1[NewsAPI + RSS] --> N2[Qwen3 NLP Agent<br/>intent classification]
        N2 -->|confirmed| N3[SOCIAL_DISRUPTION event]
    end

    R3 & H3 & A3 & M3 & N3 --> VE[Two-Step Verification]
Loading

All five triggers expose mock input controls in the Simulation Screen, allowing any parameter value to be entered manually so the full pipeline can be demonstrated live.


7.4 Two-Step Event Verification (Agentic AI)

No single data source ever directly triggers a payout. Every event must pass through two independent AI agents before reaching the claims engine.

sequenceDiagram
    participant TR as Trigger Engine
    participant A1 as Agent 1 (Structural Validator)
    participant A2 as Agent 2 (Cross-Source Corroborator)
    participant EP as Event Publisher

    TR->>A1: {source, type, ward, timestamp, raw_data}
    Note over A1: Checks: source credibility score,<br/>data completeness, geo-tag plausibility,<br/>time-series consistency
    A1->>A2: {event_candidate, confidence_1}
    Note over A2: Checks: ≥2 independent sources,<br/>LSTM pattern match,<br/>platform signal cross-check
    alt Confidence >= 85%
        A2->>EP: CONFIRMED — claim trigger + worker alert
    else 60 to 84%
        A2->>EP: WATCH — notify only, no payout
    else Below 60%
        A2->>EP: DISMISSED — log for model retraining
    end
Loading

Powered by: Qwen3 + GPT-OSS (open-source LLMs — cost-efficient and self-hostable)


7.5 Parametric Claim Engine

loss  = declared_hourly_income × disruption_hours
payout = min(loss, 5 × weekly_premium)

The claim engine (claimProcessor.ts) subscribes to confirmed disruption events, queries active workers via the zone query API, runs fraud and anti-spoofing checks, and dispatches payouts. No manual claim filing at any point.

Zero-touch settlement pipeline:

Step System Action
1. Trigger confirmed Ward-level threshold crossed, two-agent verification passes
2. Worker eligibility Active policy check, zone confirmation, no duplicate claim
3. Payout calculated loss = hourly_income × disruption_hours, cap enforced
4. Transfer initiated UPI via Razorpay Sandbox; IMPS fallback if UPI unlinked
5. Record updated Claim logged, admin dashboard loss ratio recalculated

7.6 Fraud Detection System

Rule-Based Guards (synchronous — run first):

Guard Check
Zone match Worker's GPS must be within the disruption zone polygon
Activity check Platform confirms worker was NOT completing deliveries during window
Duplicate check No overlapping claims for the same time window
Status match Platform online/offline status matches claimed disruption window

Isolation Forest (asynchronous ML layer):

Features: claim frequency vs peer group, payout/income ratio, zone vs 30-day historical location, device fingerprint consistency, time-of-claim pattern

Score Decision
< 0.3 Auto-approve, instant payout
0.3–0.6 24hr hold, human review queue
> 0.6 Reject + flag account

8. Adversarial Defense & Anti-Spoofing Strategy

8.1 The GPS Spoofing Threat Model

A fraudulent worker could attempt to:

  1. Fake location — spoof GPS to appear in a disrupted zone while working normally elsewhere
  2. Replay a legitimate signal — reuse a captured GPS trace from a valid disruption event
  3. Coordinate fraud — multiple workers submitting identical GPS traces for the same event
  4. Use mock location apps — Android developer options or third-party GPS spoofer tools

8.2 Detection Pipeline

flowchart TD
    A[Claim with GPS Data] --> B[Layer 1: Device Integrity Check]
    B --> C{Mock location app active?}
    C -->|YES| D[Instant Reject — Flag Account]
    C -->|NO| E[Layer 2: Trajectory Plausibility]
    E --> F{Speed between pings plausible?}
    F -->|Impossible| G[Reject — Teleportation detected]
    F -->|Plausible| H[Layer 3: GPS Drift Pattern]
    H --> I{Signal variance natural?}
    I -->|Zero variance — spoofed| J[Flag — Soft Hold]
    I -->|Natural| K[Layer 4: RF Environment Fingerprint]
    K --> L{Cell towers / Wi-Fi BSSIDs<br/>match GPS location?}
    L -->|Mismatch| M[Reject — RF spoofing detected]
    L -->|Match| N[Layer 5: Platform Cross-Reference]
    N --> O{Matches delivery app<br/>last-known location?}
    O -->|Mismatch > 500m| P[Soft Flag — Manual Review]
    O -->|Match| Q[Layer 6–7: Isolation Forest Score]
    Q --> R{Anomaly score?}
    R -->|< 0.3| S[Clean — Auto Approve]
    R -->|0.3–0.6| T[Hold — Manual Review]
    R -->|> 0.6| U[Reject + Flag + Notify]
Loading

8.3 Full Anti-Spoofing Check Table

Check Method Spoofing Signal Detected
Mock Location Detection Android isMockLocationEnabled flag GPS spoofer app is active
Trajectory Velocity Speed = distance(ping_n, ping_n+1) / time_delta Speed > 200km/h = physically impossible
GPS Drift Variance Statistical variance of consecutive coordinate pings Spoofed signals show unnaturally zero variance
RF Environment Fingerprint Cross-check GPS against local Wi-Fi BSSIDs + cell tower IDs GPS says Koramangala but device sees Whitefield towers
Platform Cross-Reference GigWire GPS vs delivery app's last-known location log Coordinate mismatch > 500m
Historical Zone Affinity Claimed zone vs 30-day historical delivery zones First-ever claim in a high-risk zone = elevated risk score
Peer Cluster Detection Multiple workers, identical GPS traces, same event window Coordinated spoofing pattern flagged by Isolation Forest

RF Environment Fingerprinting is the most robust layer. Even if a worker successfully spoofs their GPS coordinates, their device's Wi-Fi and cellular radio environment reflects actual physical location. A mismatch between the GPS claim and the surrounding RF infrastructure is a near-definitive spoofing signal.


9. Basis Risk Mitigation — Ward-Level Data Architecture

Basis risk is the core reason parametric insurance programs fail in India — the discrepancy between the index trigger and the actual loss on the ground. The Tata AIG program in Nagaland failed to pay a single valid claim during actual floods because trigger thresholds were built on coarse satellite data rather than local ground readings.

GigWire addresses this explicitly through a ward-level data architecture:

Approach Traditional Products GigWire
Data resolution City-level or district averages Ward / pincode level
Trigger calibration Fixed hardcoded thresholds Recalibrated weekly per pincode against 90-day rolling baseline
Weather source Satellite data or airport stations IMD API + OpenWeatherMap at street-level resolution
AQI source State-level averages CPCB ward-level monitoring station data
Validation Single source Two-agent cross-source corroboration required

By anchoring every trigger, premium, and payout to the worker's specific pincode and active delivery zone, GigWire ensures that a worker who cannot work due to localised flooding in their zone receives a payout — even if the neighbouring district registers no disruption at all.


10. Key Innovation: Privacy-First API Integration

GigWire never receives the full worker database. When a disruption is confirmed, GigWire sends a minimal zone query to the delivery app:

POST /gigwire/active-workers
{
  "zone_polygon": [...coordinates...],
  "start_timestamp": "2025-07-15T10:00:00Z",
  "end_timestamp": "2025-07-15T14:00:00Z"
}

Delivery app returns only:

[{ "worker_id": "anon_xyz_123", "active_since": "09:45", "declared_hourly_income": 90 }]

No names, phone numbers, full location history, or bank details ever leave the platform. The delivery app retains complete data sovereignty. Payouts flow through the app's in-app wallet — GigWire never touches worker bank details directly.

Delivered as a REST API and a lightweight embedded SDK (<200KB) that plugs into the delivery app's existing backend with no architectural changes required.


11. Predictive Alert & Reward System

An LSTM model produces 48-hour zone-level disruption forecasts. Workers receive push alerts via FCM when disruption probability exceeds 65%.

If a worker acts on the alert — avoids the zone or logs off — and the disruption is later confirmed, they earn a SafeDecision premium discount the following week.

Consecutive Safe Decisions Premium Discount Badge
1 −5% Rain-Ready
3 −10% Storm Shield
5 −15% Safety Champion
10 lifetime −20% permanent GigGuardian

12. Application Screens

Screen 1 — Onboarding / Registration Platform selection, city/pincode, daily earnings declaration, UPI setup. Displays the full Standard Exclusions list which the worker must acknowledge before coverage activates.

Screen 2 — Worker Dashboard Active policy status, premium paid, live zone disruption alert, earnings protected this week. Includes the What-If Panel — two side-by-side cards showing earnings recovered WITH GigWire vs total loss WITHOUT, populated with real disruption data from the current week. Dynamically updates via PayoutContext when any of the 5 triggers fires.

Screen 3 — Policy Management Weekly premium breakdown with formula shown as actual numbers. Coverage period Mon–Sun, one-tap renewal, remaining coverage cap. Explicit standard exclusions section — each exclusion listed with a brief explanation, IRDAI-compliant language.

Screen 4 — Claims & Payouts Auto-populated event log (zero manual filing). Per-event claim card: trigger type, hours lost, income lost, payout amount, UPI credit status. Full historical claim table.

Screen 5 — Insights & Savings Total earnings protected, "lost without insurance" counter, premium vs payout ROI chart, month-wise disruption vs payout bar chart, streak tracker, SafeDecision badges earned.

Screen 6 — Simulation Screen Interactive demo tool for evaluators. See Section 13.

Screen 7 — Admin / Insurer Dashboard Policy count, premiums collected, BCR / loss ratio gauge, zone-wise disruption heatmap (Mapbox), fraud flag queue with Isolation Forest scores, LSTM next-week high-claim zone forecast.


13. Admin Dashboard — Insurer Web Portal

The GigWire Admin Dashboard is a separate, standalone web application built for insurance company administrators, underwriters, and operations teams. It provides a real-time operational view of the entire GigWire platform — from enrollment metrics and premium collection to fraud queues and actuarial health indicators.

Overview

The dashboard is a Next.js web application that mirrors the visual language of the GigWire worker app — using the same colour palette, typography, and design tokens — while adapting the layout and information density for a desktop-first insurer workflow. It features fluid animations, vibrant status indicators, and a clean, professional UI theme appropriate for a regulated insurance operations context.

Key Dashboard Panels

Panel Description
User Metrics Total registered workers, new enrollments this week, worker tier breakdown (Elite / Reliable / Standard / At-Risk)
Active Policies Live count of active policies by zone and city, policies expiring this week, renewal rate
Research & Claims Open investigations, claims under active fraud review, Isolation Forest score distribution across pending claims
Premium & BCR Health Total premiums collected (current week and cumulative), BCR gauge per zone pool, loss ratio trend chart
Disruption Heatmap Mapbox-powered ward-level heatmap showing current and historical disruption events across all active zones
Fraud Flag Queue Table of flagged claims with Isolation Forest anomaly scores, GPS spoof detection flags, and one-click review actions
LSTM Forecast Panel Next-week high-claim zone predictions from the disruption forecasting model, with confidence bands
Payout Ledger Chronological log of all triggered payouts — event type, zone, worker tier, payout amount, UPI status

Mock Data & Demo Mode

For the hackathon submission, the dashboard runs entirely on mock data that simulates a realistic insurance operations environment:

  • 1,240 total registered workers across 6 cities (Bengaluru, Mumbai, Chennai, Delhi, Hyderabad, Pune)
  • 847 currently active policies (68% activation rate)
  • 23 workers flagged in the fraud review queue with varying Isolation Forest scores
  • Weekly BCR of 0.61 across the aggregate pool — within the 0.55–0.70 target range
  • 4 zone disruption events in the past 7 days — 3 confirmed payouts, 1 dismissed
  • LSTM forecast: Bengaluru South and Mumbai Dharavi rated HIGH (>65% probability) for disruption in the next 7 days

UI & Design

The admin dashboard is built to feel like a professional insurance SaaS product:

  • Theme: Dark-accented sidebar with a light main canvas — matches the GigWire app's primary brand colours (deep teal, amber accent, white text)
  • Animations: Smooth page transitions, animated gauge fills for BCR and loss ratio, live-updating counters on the metrics cards, and staggered card entrance animations on load
  • Vibrant indicators: Colour-coded status badges (green = healthy, amber = watch, red = breach) on every metric that has a defined threshold
  • Data visualisations: Recharts-powered bar charts, line charts, and gauge components for all time-series and ratio data
  • Responsive layout: Optimised for 1440px+ desktop; gracefully degrades to tablet

Running the Admin Dashboard

# From the project root
cd admin-dashboard
npm install
npm run dev
# Open http://localhost:3000

The admin portal runs independently of the main React Native / Expo app and requires no backend connection — all data is served from mock fixtures in /data/mockData.ts.


14. Simulation Screen — Live Verification Demo

The Simulation Screen demonstrates the full pipeline in real time for evaluators.

Five Interactive Trigger Scenarios Each trigger exposes mock input fields (e.g., rainfall in mm/hr, temperature in °C, AQI value). Below threshold — nothing fires. Above threshold — the full pipeline executes with live logs.

Algorithm Inspector Log Panel A terminal-style view printing real-time status as each stage processes: trigger evaluation, Agent 1 structural check, Agent 2 corroboration, confidence score outcome, GPS anti-spoofing layers, Isolation Forest fraud score, payout calculation, UPI status.

Under the Hood Dropdown Post-simulation expandable section explaining each algorithm used, why each check exists, and what attack vector each guard prevents — written for judge-friendly review.

GPS Spoofing Attack Demo A dedicated scenario simulating a fraudulent claim attempt. Shows each of the 7 anti-spoofing layers firing and demonstrates instant rejection with account flagging.

Example clean claim log:

[TRIGGER] Heavy Rain — 72mm/hr detected in Ward: Koramangala-4
[AGENT 1] Structural validation... source credibility: 0.94 PASS
[AGENT 2] Cross-source corroboration... 3 sources confirmed PASS
[CONFIDENCE] 0.91 — STATUS: CONFIRMED
[ZONE QUERY] Querying active workers in zone polygon...
[WORKER] Worker #GW-4421 — active since 09:15, Rs.90/hr declared
[GPS CHECK] Mock location flag: FALSE PASS
[GPS CHECK] Trajectory velocity: 18km/h PASS (plausible)
[GPS CHECK] Drift variance: 0.0042 PASS (natural signal)
[GPS CHECK] RF fingerprint: cell towers match GPS location PASS
[GPS CHECK] Platform cross-reference: within 120m PASS
[ISOLATION FOREST] Anomaly score: 0.14 — CLEAN
[CLAIM] Loss = Rs.90 x 4hrs = Rs.360 | Coverage cap: Rs.175 | Payout: Rs.175
[PAYOUT] UPI credit initiated via Razorpay Sandbox PASS

Example blocked spoofing attempt:

[GPS CHECK] Mock location flag: TRUE FAIL
[GPS CHECK] RF fingerprint: GPS says Koramangala, towers show Whitefield FAIL
[DECISION] INSTANT REJECT — GPS spoofing detected
[ACTION] Account flagged for review. Worker notified.

Home Screen Sync When any trigger fires in the simulation, the Home Screen dashboard updates live via PayoutContext — earnings protected, active alert banner, and payout history all reflect the simulated event.


15. Tech Stack

Application

Layer Technology
Mobile App React Native + Expo (web-runnable via npm run web)
Admin Panel Next.js
Offline Storage AsyncStorage + SQLite
Charts Victory Native
Push Notifications Firebase Cloud Messaging (FCM)
Maps / Geofencing Mapbox

Backend & Services

Layer Technology
Core API Node.js + NestJS
Database PostgreSQL + Supabase
Cache / Queue Redis + Bull
ML Service Python + FastAPI + scikit-learn + XGBoost
AI / LLM Agents Qwen3 + GPT-OSS
Event Streaming Apache Kafka

External APIs

API Purpose
OpenWeatherMap + Meteomatics Weather triggers — rain, heat (ward-level)
CPCB AQI API Air quality trigger (ward-level monitoring stations)
Mapbox Zone polygon matching, traffic trigger, disruption heatmap
NewsAPI + Local RSS Social disruption NLP trigger
NDMA / State Alerts Government event feeds
Razorpay Sandbox Simulated UPI payout processing
NPCI UPI API UPI credit flow simulation

16. Platform Strategy — Mobile App + Embedded SDK

Channel A — GigWire App (Standalone) React Native + Expo, web-runnable at http://localhost:8081. Target: Android 8+, <15MB APK, 2G/3G optimised, offline-first with local SQLite caching. For workers on any delivery platform.

Channel B — GigWire SDK (Embedded) Node.js npm package, <200KB. Integrates into the delivery app's existing backend — stateless, no architectural changes required, queue-based retry for network failures. Zone query API + webhook event notifier. Platform retains full data control.


17. Weekly Pricing Model

Policy Period:    Monday 00:00 to Sunday 23:59
Premium Charged:  Sunday midnight (auto-debit or one-tap renewal)
Payout Window:    Within 10 minutes of event verification
Coverage Cap:     5x weekly premium (max payout per week)
BCR Target:       0.55 to 0.70 (monitored per zone pool)

Sample premiums by city and season:

Worker Profile Zone Risk Season Factor GScore Weekly Premium
Mumbai, July (monsoon), Elite 1.60 1.80 0.85 ~₹86
Bangalore, March (dry), Standard 1.00 0.80 0.50 ~₹28
Chennai, December (cyclone), Reliable 1.45 1.60 0.70 ~₹57
Delhi, November (severe AQI), New worker 1.35 1.40 0.22 ~₹66

18. Data Flow Diagrams

End-to-End Zero-Touch Claim Flow

flowchart TD
    A([Worker registers — exclusions acknowledged]) --> B[GScore calculated]
    B --> C[Weekly premium quoted — ward-level risk]
    C --> D[Policy active: Mon–Sun]
    D --> E{Any of 5 triggers fires?}
    E -->|No| F[Week ends — renewal prompt]
    E -->|Yes| G[Two-agent verification pipeline]
    G --> H{Confidence >= 85%?}
    H -->|No| I[WATCH or DISMISSED — logged]
    H -->|Yes| J[Zone query to delivery app]
    J --> K[Active workers in ward identified]
    K --> L[GPS anti-spoofing — 7-layer check incl. RF fingerprint]
    L --> M[Isolation Forest fraud score]
    M --> N{Score < 0.3?}
    N -->|Yes| O[Auto-approve payout via Razorpay]
    N -->|No| P[Hold or Reject]
    O --> Q[Worker dashboard updates live]
    O --> R[Admin: BCR and loss ratio recalculated]
    Q --> S[SafeDecision reward evaluated]
    S --> T[Next week premium adjusted]
Loading

19. Research References

  1. Gohil, J. & Jha, A. (2024). _ Addressing Policy Gaps for Gig Workers in India: A Focus on Food Delivery Platforms. _ IJFMR, Vol. 6, Issue 6. E-ISSN: 2582-2160. — GScore formula, Section 4.2

  2. NITI Aayog (2024–2025). Gig and Platform Workers in India — workforce size, growth projections, and protection gap data.

  3. Sabka Bima Sabki Raksha Act (2025). Amendment to Insurance Act 1938 and IRDAI Act 1999 — gig worker coverage mandate and micro-insurance national priority declaration. KPMG First Notes, January 2026.

  4. IRDAI Insurance Products Regulations (2023). Micro-insurance product guidelines, transparency requirements, and sandbox framework for parametric InsurTech.

  5. Parametric Insurance Literature — Burning Cost Rate methodology, basis risk management, reinsurance tower design. Swiss Re Corporate Solutions; Artemis.bm; Wharton Impact.

  6. OpenWeatherMap API — rainfall and temperature thresholds at ward level

  7. CPCB — AQI classification standards (Hazardous: >300)

  8. IMD — Colour-coded weather alert system (Red Alert classification)

  9. NDMA — Disaster alert classification for parametric trigger design

  10. Mapbox — Zone polygon geofencing and traffic velocity thresholds


Built for India's 12 million gig workers — and the 23.5 million more on the way.

_ GigWire — Seed. Scale. Soar. _

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors