Skip to content

techstar9797/sentinel-guard

BranchesTags

Repository files navigation

Sentinel - AI-Powered Fraud Detection Platform

Sentinel is a privacy-first, AI-powered fraud detection platform that combines blockchain intelligence, behavioral analytics, and continuous learning to protect digital assets while maintaining regulatory compliance.

🏗️ Architecture Overview

Multi-Agent System

Sentinel employs a sophisticated multi-agent architecture where specialized AI agents work collaboratively:

  1. Watcher Agent (/)

    • Real-time transaction monitoring dashboard
    • Live activity feed and case overview
    • Key metrics visualization (total cases, high priority alerts, detection rate)
    • Quick access to recent investigations

  2. Detective Agent (/cases, /investigation)

    • Deep investigation capabilities
    • Risk analysis and evidence gathering
    • Integration with TRM Labs for blockchain intelligence
    • Moralis API for comprehensive transaction data
    • Case management and status tracking

  3. Guardian Agent (/guidelines)

    • Decision framework enforcement
    • Risk-based approval/decline logic
    • Compliance rule management
    • Automated decision recommendations

  4. Coach Agent (/playbooks, /performance)

    • Pattern recognition and playbook generation
    • Continuous learning from case outcomes
    • Performance metrics tracking (APS - Agent Performance Score)
    • False positive/negative rate optimization
    • Agent version management and A/B testing

🔒 Privacy-First Architecture

Skyflow Integration

Sentinel implements a zero-trust PII architecture using Skyflow's Data Privacy Vault:

┌─────────────────────────────────────────────────────────────┐
│                      Sentinel Platform                        │
├─────────────────────────────────────────────────────────────┤
│                                                               │
│  ┌──────────────┐                      ┌─────────────────┐  │
│  │  Frontend    │◄────────────────────►│  Fraud Engine   │  │
│  │  (React)     │   Tokenized Data     │  (Edge Funcs)   │  │
│  └──────────────┘                      └─────────────────┘  │
│         │                                        │           │
│         │                                        │           │
│         ▼                                        ▼           │
│  ┌────────────────────────────────────────────────────┐     │
│  │         Supabase Database (Tokens Only)            │     │
│  │  - wallet_address                                  │     │
│  │  - name_token (sky_xxxxx)                         │     │
│  │  - email_token (sky_xxxxx)                        │     │
│  │  - phone_token (sky_xxxxx)                        │     │
│  │  - address_token (sky_xxxxx)                      │     │
│  └────────────────────────────────────────────────────┘     │
│                           │                                  │
└───────────────────────────┼──────────────────────────────────┘
                            │
                            │ Tokenization/Detokenization
                            │ (AES-256 Encryption)
                            ▼
                 ┌──────────────────────┐
                 │   Skyflow Vault      │
                 │   (PII Storage)      │
                 │                      │
                 │  - Actual Names      │
                 │  - Email Addresses   │
                 │  - Phone Numbers     │
                 │  - Addresses         │
                 │                      │
                 │  Encrypted at Rest   │
                 │  & in Transit        │
                 └──────────────────────┘

Key Privacy Features:

  • No PII in Application Database: Only Skyflow tokens are stored
  • Tokenization at Entry: PII is tokenized before reaching the application
  • Audited Detokenization: All PII access is logged and requires justification
  • GDPR & CCPA Compliant: Right to be forgotten, data minimization
  • Data Residency: Configurable vault locations for regional compliance

Compliance Dashboard

The compliance dashboard (/compliance) provides real-time visibility into:

  • PII tokenization coverage (100% target)
  • GDPR/CCPA compliance status
  • Detokenization request audit trail
  • Data retention policy enforcement
  • Analyst access monitoring

🔍 Decision Guidelines

Sentinel implements risk-based decision logic:

Approval Criteria

  • ✅ TRM Risk Score: < 0.5 (Low Risk)
  • ✅ Dark Web Signals: None detected
  • ✅ Transaction Velocity: Within normal patterns
  • ✅ Geo-Location: Not from sanctioned regions

Decline Criteria

  • ❌ TRM Risk Score: ≥ 0.9 (High Risk)
  • ❌ Sanctioned Entity: OFAC/EU sanctions match
  • ❌ Dark Web Activity: Confirmed marketplace transactions
  • ❌ Ransomware Association: Direct or indirect link

Manual Review Criteria

  • 🔍 TRM Risk Score: 0.5 - 0.9 (Medium Risk)
  • 🔍 Unusual Patterns: Deviates from user history
  • 🔍 New User: First transaction with limited history
  • 🔍 High Value: Above configured threshold

🧠 Continuous Learning Cycle

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   Monitor    │────►│   Detect     │────►│   Decide     │
│  (Watcher)   │     │ (Detective)  │     │ (Guardian)   │
└──────────────┘     └──────────────┘     └──────────────┘
       ▲                                          │
       │                                          ▼
       │                                   ┌──────────────┐
       │                                   │   Execute    │
       │                                   │  (Action)    │
       │                                   └──────────────┘
       │                                          │
       │              ┌──────────────┐           │
       └──────────────│    Learn     │◄──────────┘
                      │   (Coach)    │
                      └──────────────┘

The Coach agent continuously:

  1. Analyzes case outcomes (approved, declined, false positives)
  2. Identifies emerging fraud patterns
  3. Generates new playbooks automatically
  4. Adjusts risk thresholds based on performance metrics
  5. Suggests agent parameter tuning

🛠️ Technology Stack

Frontend

  • React 18 with TypeScript
  • Vite for fast development and building
  • TailwindCSS for styling with custom design system
  • Shadcn UI components
  • React Query for data fetching and caching
  • React Router for navigation

Backend (Lovable Cloud)

  • Supabase for database and authentication
  • Edge Functions for serverless business logic
  • PostgreSQL with Row Level Security (RLS)

Integrations

  • TRM Labs: Blockchain intelligence and risk scoring
  • Moralis: Multi-chain transaction data and wallet analytics
  • Skyflow: PII tokenization and data privacy vault
  • Anthropic Claude: AI-powered decision reasoning (via edge functions)
  • AWS/Redis: Caching and performance optimization

📁 Project Structure

sentinel/
├── src/
│   ├── components/          # Reusable UI components
│   │   ├── ui/             # Shadcn components
│   │   ├── DecisionBadge.tsx
│   │   ├── RiskBadge.tsx
│   │   ├── Layout.tsx
│   │   └── Sidebar.tsx
│   ├── pages/              # Route components
│   │   ├── Index.tsx       # Watcher dashboard
│   │   ├── Cases.tsx       # Case list
│   │   ├── Investigation.tsx
│   │   ├── Guidelines.tsx
│   │   ├── Playbooks.tsx
│   │   ├── Performance.tsx
│   │   ├── Compliance.tsx
│   │   └── DemoPrivacy.tsx
│   ├── services/           # API integrations
│   │   ├── trmService.ts
│   │   └── moralisService.ts
│   ├── data/              # Mock data for demo
│   └── types/             # TypeScript definitions
├── supabase/
│   ├── functions/         # Edge functions
│   │   ├── trm-screen/
│   │   ├── moralis-screen/
│   │   ├── skyflow-tokenize/
│   │   ├── skyflow-detokenize/
│   │   └── screen-with-privacy/
│   └── migrations/        # Database schema
└── public/
    └── assets/logos/      # Partner logos (TRM, Skyflow, etc.)

🚀 Getting Started

Prerequisites

  • Node.js 18+ and npm
  • Supabase account (via Lovable Cloud)
  • API keys for integrations:
    • TRM Labs API key
    • Moralis API key
    • Skyflow API key
    • Redis credentials (optional, for caching)

Installation

  1. Clone the repository

    git clone https://github.com/techstar9797/sentinel-guard.git
cd sentinel-guard

  2. Install dependencies

    npm install

  3. Configure environment variables

    Environment variables are managed through Supabase secrets:

    • SUPABASE_URL and SUPABASE_ANON_KEY (auto-configured)
    • TRM_API_KEY - TRM Labs API key
    • MORALIS_API_KEY - Moralis API key
    • SKYFLOW_API_KEY - Skyflow vault API key
    • REDIS_ACCOUNT_KEY, REDIS_USER_KEY (optional)

  4. Start development server

    npm run dev

  5. Access the application

    Open http://localhost:5173

📊 Database Schema

Cases Table

cases (
  id UUID PRIMARY KEY,
  case_number TEXT UNIQUE,
  wallet_address TEXT,
  name_token TEXT,          -- Skyflow token
  email_token TEXT,         -- Skyflow token
  phone_token TEXT,         -- Skyflow token
  address_token TEXT,       -- Skyflow token
  status TEXT,              -- pending, investigating, resolved
  priority TEXT,            -- low, medium, high, critical
  decision TEXT,            -- approved, declined, manual_review
  trm_risk_score FLOAT,
  trm_risk_level TEXT,
  trm_evidence JSONB,
  agent_version TEXT,
  created_at TIMESTAMP,
  updated_at TIMESTAMP
)

Compliance Metrics Table

compliance_metrics (
  id UUID PRIMARY KEY,
  metric_date DATE,
  gdpr_compliant BOOLEAN,
  ccpa_compliant BOOLEAN,
  data_residency TEXT,
  tokenized_fields INTEGER,
  total_pii_fields INTEGER,
  tokenization_percentage FLOAT,
  detokenization_requests INTEGER,
  analyst_access_count INTEGER,
  created_at TIMESTAMP
)

🔐 Security Considerations

  1. Row Level Security (RLS): All database tables have RLS policies enabled
  2. API Key Management: All keys stored as encrypted Supabase secrets
  3. Audit Logging: All detokenization requests are logged with timestamp and user
  4. Zero-Knowledge Architecture: Application never sees raw PII
  5. Rate Limiting: Edge functions implement rate limiting to prevent abuse

📈 Performance Metrics

The platform tracks key performance indicators:

  • APS (Agent Performance Score): Composite score of detection accuracy
  • False Positive Rate: Percentage of legitimate transactions flagged
  • False Negative Rate: Percentage of fraud cases missed
  • Loss Prevented: Estimated fraud amount prevented
  • Dark Web Signal Usage: Percentage of cases using dark web intelligence
  • Average Investigation Time: Time to case resolution

🔄 CI/CD

Changes pushed to the repository are automatically synced with Lovable Cloud. Edge functions are deployed automatically upon changes.

📄 License

This project is private and proprietary.

🤝 Contributing

This is a private project. For questions or issues, contact the development team.

📞 Support

For technical support or questions about the platform, please contact:

  • Technical Lead: [Your contact information]
  • Security Issues: [Security contact]

Built with ❤️ using Lovable, Supabase, TRM Labs, Skyflow, and Moralis

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages