Real-Time Income Protection for Gig Workers
Guidewire DevTrails 2026 — Phase 1 Submission
Platform: Progressive Web App
Backend: Python | FastAPI
AI/ML: XGBoost | Scikit-learn | Isolation Forest
- Title & Introduction
- Problem Statement
- Personas — Who We Protect
- Core Problem
- Solution (Overview)
- Key Features
- How It Works
- Scenario
- Premium Model
- Trigger System
- AI & ML Integration
- Adversarial Defense & Anti-Spoofing Strategy
- System Architecture
- Application Workflow
- Platform Choice (PWA)
- Tech Stack
- Future Scope
- Development Roadmap
- Closing Note
HyperNova is an AI-powered parametric insurance platform designed to protect gig workers from income loss caused by real-world disruptions.
Unlike traditional insurance systems that require manual claims and verification, HyperNova operates automatically. It continuously monitors environmental and behavioral signals, detects disruption events, estimates income loss, and triggers instant payouts.
This approach transforms insurance from a reactive system into a proactive, real-time financial safety net.
India’s Q-commerce ecosystem (Blinkit, Zepto, Swiggy Instamart) depends on gig workers who earn on a per-delivery basis.
This model creates a critical vulnerability: If the worker cannot work, they earn nothing.
Disruptions include:
- Smog (AQI above safe levels)
- Heavy rain and flooding
- Extreme heat conditions
- Safety risks during night shifts
- Platform outages or forced logoffs
These disruptions are not gradual — they cause immediate work stoppage.
As a result:
- Daily income drops to zero instantly
- Workers are forced to choose between safety and survival
- There is no fallback or compensation mechanism
"One bad day equals one week of groceries gone"
| Attribute | Details |
|---|---|
| Age | 23 |
| Work | Blinkit/Zepto, 12 hours/day |
| Earnings | ₹500–700/day |
| Annual Loss | ₹26,400 |
| Biggest Fear | AQI 400 means zero income |
Arjun represents workers whose income is directly tied to daily activity. Environmental disruptions frequently force him to stop working, resulting in immediate financial loss.
"I earn the same as men, but I cannot work when they can"
| Attribute | Details |
|---|---|
| Age | 26 |
| Work | Zepto/Blinkit, 8 hours/day (wants 12) |
| Earnings | ₹400–500/day |
| Annual Loss | ₹62,400 |
| Biggest Fear | Losing income after 7 PM daily |
Priya represents workers impacted by safety constraints. She is unable to work during high-income time slots, leading to consistent and predictable income loss.
| Disruption | Example | Impact |
|---|---|---|
| Air Pollution | AQI > 400 | Unsafe to work |
| Rain/Flood | Roads blocked | No deliveries |
| Heat | > 45°C | Health risk |
| Traffic Curfew / Restrictions | Police restrictions, VIP movement, emergency lockdowns | Forced idle time |
| Safety Risk | Night shifts | Unsafe, especially for women |
| Platform Issues | Downtime / forced logoff | Zero income |
The problem is not reduced productivity — it is forced inactivity.
Gig workers do not slow down during disruptions.
They are forced to stop working completely due to:
- Government restrictions (curfews, traffic blocks)
- Environmental hazards (AQI, heat, rain)
- Safety risks (especially after dark)
- Platform-level failures
Unlike salaried employees:
- No work = No pay
- No compensation during disruption
- No insurance coverage for income loss
- Daily loss: ₹400–₹700
- Monthly loss: ₹3,000–₹6,000
- Annual loss: Up to 2–5 months of income
Gig workers face zero-income risk during uncontrollable disruptions,
with no system to compensate for forced inactivity.
HyperNova introduces a parametric insurance model where payouts are triggered based on predefined real-world conditions rather than manual claims.
The system performs three core functions:
- Detect disruptions using real-time external data (AQI, weather, platform signals)
- Estimate income loss using AI models based on user behavior and historical earnings
- Automatically trigger payouts when conditions are met
This removes friction, delays, and subjectivity from traditional insurance workflows.
| Feature | Description |
|---|---|
| Parametric Auto-Payout | Automatic compensation triggered by verified disruptions — no claims or manual process. |
| Hyper-Local Detection | Detects disruptions (rain, AQI, curfew, outages) at zone-level precision. |
| Weekly Micro-Insurance | Flexible ₹50–₹80 weekly plans aligned with gig workers’ income cycle. |
| Instant Payout Engine | UPI-based automated payouts processed within minutes. |
| Rollback Protection | Compensates workers for remaining expected earnings if they stop mid-shift due to disruption. |
| Premium Gifting | Users can sponsor insurance for other workers, enabling a community safety net. |
| Reward System | Consistent activity and safe behavior reduce risk score → lower premiums. |
| SOS + Safety Mode | Emergency trigger for unsafe situations, especially helpful for women workers. |
| Fraud Prevention | Movement tracking, activity validation, and anomaly detection to prevent misuse. |
| AI Risk Prediction | Predicts disruption probability for smarter pricing and early alerts. |
The system operates as a continuous loop:
- User onboarding captures basic profile and working patterns
- AI models calculate risk score and assign a premium
- System continuously monitors environmental and activity signals
- When a trigger condition is detected, the system evaluates its validity
- Expected income is calculated using historical patterns
- Actual income is compared with expected income
- Fraud detection layer validates authenticity
- If valid, payout is triggered instantly
Location: Delhi
Condition: AQI reaches 420
- Worker logs off due to unsafe air quality
- System detects AQI threshold breach
- AI calculates expected earnings for the remaining day
- Actual earnings are compared
- Loss is estimated at ₹520
- Fraud checks confirm genuine activity
- ₹520 is credited instantly
Premium = (Base × Risk Score) + (Expected Loss × Probability)
Explanation:
- Base: Minimum operational cost
- Risk Score: Derived from location, time, environmental exposure
- Expected Loss: Average income loss for similar users
- Probability: Likelihood of disruption occurring
This ensures:
- Fair pricing for users
- Sustainability of the system
- Dynamic adaptation to risk conditions
| Trigger | Threshold | Source |
|---|---|---|
| AQI | > 400 | CPCB API |
| Rain | Heavy | Weather API |
| Heat | > 45°C | Weather API |
| Safety | Manual | User input |
| Platform | Downtime | System logs |
Triggers are objective, measurable, and externally validated, ensuring transparency.
Predicts the likelihood of disruption based on environmental and behavioral patterns.
Estimates expected earnings using:
- Time of day
- Location demand
- Historical user performance
Detects anomalies such as:
- Inactive users claiming loss
- Impossible movement patterns
- Repeated identical claims
System Philosophy: From Detection to Deterrence
Most fraud systems ask: "Is this claim fake?"
We ask: "Why would anyone even try?"
Our architecture combines pre-qualification barriers, economic disincentives, and parametric automation to make fraud not just detectable, but pointless.
- Proof-of-Work Requirement Concept: Before payout, system verifies worker was actively working when disruption hit.
Requirement Data Source Why It Kills Fraud Order Logs Delivery platform API Fraudster at home = no orders GPS Trail Last 30 min movement Must show movement toward zone App Activity Usage metrics Active worker vs idle phone
Logic:
IF orders_last_60min = 0 AND app_foreground = false → CLAIM INVALID (No Proof-of-Work) Fraudster's Dilemma: "I can spoof GPS, but I cannot fake delivery work I never did."
- Zone-Level Parametric Payouts Concept: For major disruptions, we eliminate claims entirely.
Scenario Action Fraud Impact Red Alert triggered Scan all active workers in zone No claims filed Workers in zone Auto-payout 100% Nothing to fake Workers near zone (<2km) Partial payout (70%) Fairness ensured Workers outside No payout Clean cutoff Why This Works: No claims → nothing to spoof → fraud rings lose entry point
- Economic Disincentive Layer Fraud doesn't just fail — it becomes unprofitable.
Zone Fraud Level Payout Impact Community Response <5% 100% Normal 5-15% 80% + verification Self-policing 15-30% 50% + manual review Fraud becomes costly
30% Zone freeze (24h) Unprofitable to attack 👉 Genuine workers compensated via adjusted future premiums
Tier 1: Rule-Based Filters (Instant Kill) Rule Threshold Action Velocity >80 km/h Auto-Reject WiFi Clustering >5 claims/same BSSID Hold + Ring Flag Activity Proof No work last 60 min Invalid Device Integrity Rooted / Mock location High Risk + Challenge Tier 2: AI/ML Models Model Purpose Output Isolation Forest Anomaly detection 0-100 anomaly score DBSCAN Fraud ring detection Cluster density score XGBoost Trust scoring 0-100 trust score
Decision Engine Fraud Confidence = (0.4 × Anomaly) + (0.4 × RingScore) + (0.2 × (100 - Trust))
Score Action User Message 0-30 Instant Pay "Payment processed instantly" 31-60 Soft Delay "Verification in progress (≤4 hrs)" 61-85 Challenge Selfie / liveness check 86-100 Reject Soft rejection + support
Trigger Condition = TRUE (rainfall > threshold) AND Worker Eligibility = TRUE (Proof-of-Work / Zone presence) → PAYOUT TRIGGERED
No trigger = No payout | No work = No payout
To exploit, attacker must spoof ALL simultaneously:
. GPS location
. Sensor telemetry (movement, light)
. Network identity (WiFi, IP, cell towers)
. Behavioral history (delivery patterns)
Spoofing one signal = Easy Spoofing ALL signals consistently = Impossible at scale
Plus:
DBSCAN detects coordinated rings
Feedback loops adapt within 24h
Fraud becomes economically irrational, not just technically difficult
| Metric | Impact |
|---|---|
| Fraud Reduction | Near-complete elimination of low-effort fraud |
| Ring Attacks | Structurally prevented via zone payouts |
| Manual Reviews | ~80% reduction |
| User Experience | Instant payouts for trusted workers |
| System Stability | Liquidity protected under attack |
| Scenario | Problem | HyperNova Solution |
|---|---|---|
| Trapped in flood | Can't move, GPS static | Emergency override + zone payout |
| Phone dead | No app activity | Last known location + prior work history |
| Evacuated | Left zone during disaster | Partial payout for time worked |
| Curfew stuck | Can't go home, can't work | Night premium + safety bonus |
| Smog at home | Didn't go to work | No payout (no work = no coverage) |
The architecture follows a modular pipeline:
Frontend (PWA) handles user interaction
Backend (FastAPI) processes requests
AI Engine performs prediction and estimation
Trigger Engine evaluates disruption conditions
Fraud Engine validates authenticity
Payout Engine executes transactions
Database stores user and transaction data
External APIs provide real-time environmental inputs.
Onboarding → Risk Profiling → Subscription
→ Monitoring → Trigger Detection
→ Loss Calculation → Fraud Validation
→ Instant Payout
The system uses a Progressive Web App to ensure accessibility:
- Works on low-end Android devices
- No installation required
- Minimal storage usage
- Faster load times
| Layer | Technology |
|---|---|
| Backend | FastAPI |
| AI | XGBoost, Scikit-learn |
| Fraud | Isolation Forest |
| Database | MongoDB |
| Frontend | HTML, JavaScript (PWA) |
| Payments | Razorpay |
- Native Android application
- Predictive alerts before disruptions
- Integration with delivery platforms
- Expansion across cities
- AI-driven safety routing for women
| Phase | Status |
|---|---|
| Phase 1 | Concept and architecture complete |
| Phase 2 | Backend and API development |
| Phase 3 | AI model optimization and scaling |
Gig workers do not need charity.
They need predictable income protection.
HyperNova ensures that when disruption happens, income does not disappear.