Grid-Guardian

AI-powered platform that prevents solar subsidy fraud by combining Salesforce automation, Tableau Next analytics, and Slack alerts.

Comment 1

Inspiration

Solar subsidy programs are designed to accelerate clean energy adoption, but in practice they suffer from delayed verification, data silos, and fraudulent claims that are often detected only after funds are disbursed. We were inspired by the gap between policy intent and operational reality. Grid-Guardian was born from the idea that fraud detection should be real-time, automated, and actionable, not manual and retrospective.

What it does

Grid-Guardian is an end-to-end, AI-driven fraud detection platform that automatically verifies solar subsidy claims by correlating claim data, satellite verification, and power generation metrics.
It uses Salesforce for intelligent automation and decision-making, Tableau Next for fraud pattern analytics, and Slack for real-time alerts and conversational queries—creating a closed-loop fraud prevention ecosystem.

How we built it

  • Salesforce as the system of record to capture subsidy claims and manage verification workflows
  • Salesforce Flows & AI logic to calculate risk scores, classify fraud levels, and automate decisions
  • Tableau Next connected to Salesforce data for real-time dashboards, fraud trends, and impact analysis
  • Slack Bot integrated via REST APIs to deliver alerts and enable slash-command-based data access
  • Modular architecture designed to scale across regions and subsidy programs

Grid-Guardian Architecture (Formula-Based on Semantic Modelling)

1. Claim Ingestion Layer

Purpose: Capture subsidy claims at source

Components

  • Salesforce LWC Dashboard
  • External systems via REST APIs

Captured Data

  • Reported panel count
  • Reported capacity (kW)
  • Installer metadata
  • Claim location

This layer provides the raw inputs for downstream verification and risk calculations.

2. Salesforce – System of Record

Purpose: Centralized and trusted data storage

Core Objects

  • Subsidy_Claim__c
  • Satellite_Verification__c
  • Generation_Data__c
  • Investigation_Case__c

Characteristics

  • Auditable
  • Policy-compliant
  • Single source of truth

All calculations operate on validated Salesforce data.

3. AI-Driven Risk Intelligence

Purpose: Detect fraud signals before subsidy disbursement

Panel Mismatch Detection

$$ \text{Panel Mismatch} = \frac{|\text{Reported Panels} - \text{AI Detected Panels}|} {\text{Reported Panels}} \times 100 $$

Generation Efficiency Deviation

$$ \text{Efficiency Loss} = \left(1 - \frac{\text{Actual Generation (kWh)}} {\text{Theoretical Max Generation (kWh)}}\right) \times 100 $$

Capacity Plausibility Validation

$$ \text{Plausibility} = \min\left( 100,\ \frac{\text{Reported Capacity (kW)}} {\text{Roof Area (sqft)} \times 0.015} \right) $$

Overall Verification Score

Verification Score = 0.4 × (100 − Panel Mismatch) + 0.4 × Satellite Confidence + 0.2 × Plausibility

Final Risk Score

$$ \text{Risk Score} = 100 - \text{Verification Score} $$

4. Decision Orchestration Layer

Purpose: Convert intelligence into deterministic decisions

Mechanisms

  • Salesforce Flows
  • Agentforce Fraud Analyst Agent

$$ \text{Decision} = \begin{cases} \text{APPROVED} & \text{if Risk Score} < 30 \land \text{Verification Score} > 70 \ \text{MANUAL REVIEW} & 30 \le \text{Risk Score} < 70 \ \text{REJECTED} & \text{if Risk Score} \ge 70 \end{cases} $$

Persisted Fields

  • Risk_Score__c
  • Risk_Level__c
  • AI_Analysis__c
  • Final_Decision__c

5. Analytics & Pattern Discovery

Purpose: Strategic fraud analysis and policy insight

Tool

  • Tableau Next

Insights

  • Fraud trends
  • Regional risk heatmaps
  • Installer anomaly patterns
  • Subsidy leakage and savings impact

6. Real-Time Action & Alerting

Purpose: Immediate operational response

Integration

  • Slack Bot: SlackAnalyticsBot

Capabilities

  • Real-time fraud alerts
  • Conversational analytics

Commands

  • /gridguardian stats
  • /gridguardian fraud-today
  • /gridguardian pending

7. Governance & Trust

Purpose: Transparency, security, and scalability

Controls

  • Full audit trail in Salesforce
  • Explainable AI outputs
  • Policy-aligned decision thresholds
  • Scalable across regions and subsidy programs

What we learned

  • Fraud prevention is most effective when analytics and action are tightly coupled
  • Dashboards alone don’t stop fraud—real-time delivery to decision-makers does
  • Low-code platforms like Salesforce can power complex, AI-driven workflows at scale
  • Clear explanations of AI decisions are as important as the decisions themselves

What's next for Grid-Guardian

  • Integrating advanced computer vision for satellite image analysis
  • Adding adaptive learning to improve fraud detection over time
  • Expanding Slack workflows to support approvals and escalations
  • Scaling the platform to other subsidy, insurance, and compliance use cases

Built With

Share this project:

Updates