Sentinel-Service-Architecture

Sentinel is AIGuardian's comprehensive AI Security & Safety protection service that provides multi-tenant SaaS capabilities for integrating Input and Output Guardrails. It detects, quantifies, and mitigates risks like prompt injection, toxicity, and PII leakage across generative AI applications.

Core Architecture Components

Sentinel Services

• sentinel-web (Next.js Frontend)

  • Analytics Dashboard for monitoring and insights
  • Sentinel Playground for ad-hoc testing
  • Multi-tenant web interface with JWT authentication

• sentinel-api (FastAPI Backend)

  • Primary validation API for external integrations
  • Handles guardrail execution and scoring
  • Supports parallel processing of multiple guardrails

• sentinel-core-api (NestJS Core Logic)

  • Internal analytics and data processing
  • Background task management
  • Data export and reporting functionality

• sentinel-db (Prisma Models Package)

  • Database abstraction layer
  • Multi-tenant data isolation
  • PostgreSQL schema management

System Architecture Diagram

graph TB
    %% External Users & Systems
    subgraph "External Access"
        Users[👥 Tenant Users<br/>Web Dashboard]
        Litmus[🧪 Litmus Service<br/>Testing Integration]
    end

    %% AWS Infrastructure Layer
    subgraph "AWS Infrastructure"
        subgraph "Public Gateway"
            GATEWAY[🌐 Public Gateway<br/>WAF + CloudFront + API Gateway<br/>API Key Validation]
        end

        subgraph "EKS Cluster"
            PROXY[🦍 Load Balancer + Kong<br/>Internal Routing & Load Balancing]

            subgraph "Sentinel Services"
                SWEB[🌐 sentinel-web<br/>Next.js Frontend]
                SAPI[🔧 sentinel-api<br/>FastAPI Backend]
                SCORE[⚙️ sentinel-core-api<br/>NestJS Core Logic]
            end
        end

        subgraph "Data Storage"
            SENTINELDB[(🗄️ Sentinel Database<br/>PostgreSQL RDS<br/>Guardrails & Check Results)]
            STORAGE[� Data Storage<br/>PostgreSQL RDS + S3<br/>Multi-AZ & Backup]
        end
    end

    %% AIGuardian Ecosystem
    subgraph "AIGuardian Ecosystem"
        AUTH[🔐 Authentication System<br/>UAM Service + Database<br/>Users, Tenants, API Keys]
    end

    %% All Guardrail Services Combined
    subgraph "Guardrail Services"
        EXTERNAL[☁️ External Services<br/>AWS Bedrock + Azure AI<br/>OpenAI + Cloak]
        CUSTOM[🤖 Custom AI Models<br/>LionGuard + PromptGuard<br/>Off-Topic + Leak Detection]
    end

    %% Simplified Data Flow Connections
    Users -->|HTTPS + JWT| GATEWAY
    Litmus -->|Internal API| SAPI

    GATEWAY --> PROXY
    PROXY --> SWEB
    PROXY --> SAPI

    SWEB -->|Internal API| SCORE
    SAPI -->|Background Tasks| SCORE

    %% Authentication Flow
    SWEB -.->|JWT Validation| AUTH
    SAPI -.->|API Key Validation| AUTH
    SCORE -.->|Auth Guards| AUTH

    %% Database Connections
    SWEB --> SENTINELDB
    SAPI --> SENTINELDB
    SCORE --> SENTINELDB

    %% Guardrail Integrations
    SAPI -->|Parallel Validation| EXTERNAL
    SAPI -->|Model Inference| CUSTOM

    %% Data Export & Storage
    SCORE -->|Export Jobs| STORAGE

External Integrations

AIGuardian Ecosystem

• UAM Service: Centralized authentication and authorization
• Multi-tenant Database: PostgreSQL with tenant isolation

External Guardrail Services

• AWS Bedrock Guardrails: Content filtering and safety checks
• Azure AI Content Safety: Microsoft's content moderation
• OpenAI Moderation API: OpenAI's content filtering
• Cloak Service: Additional privacy protection

Custom AI Models

• LionGuard: GovTech-developed toxicity detection
• PromptGuard: Meta's prompt injection detection
• Off-Topic Detection: GovTech model for relevance checking
• System Prompt Leakage: Detection of prompt extraction attempts

Infrastructure Dependencies

AWS Services

• EKS Cluster: Container orchestration platform
• RDS PostgreSQL: Multi-AZ database with high availability
• S3 Buckets: Data export and file storage
• CloudFront: CDN for global content delivery
• API Gateway: API key validation and rate limiting
• WAF: Web application firewall protection

Container Architecture

• Kong Proxy: Internal service routing and load balancing
• Application Load Balancer: Traffic distribution
• Docker Containers: Microservices deployment
• Kubernetes: Container orchestration and scaling

Key Architectural Patterns

1. Security-First Design

• Multiple validation layers (WAF → API Gateway → Kong → Services)
• API key validation at gateway level
• JWT token authentication for web users
• Tenant isolation through headers and database design

2. Microservices Architecture

• Separate concerns between web, API, and core services
• Independent scaling and deployment
• Service-specific databases and configurations

3. Asynchronous Processing

• Background tasks for data storage
• Non-blocking guardrail execution
• Export job scheduling and processing

4. Multi-Cloud Integration

• Support for AWS, Azure, and OpenAI services
• Vendor-agnostic guardrail framework
• Flexible model integration patterns

5. Monitoring & Analytics

• Real-time validation tracking
• Historical data analysis
• Export capabilities for compliance reporting

This architecture documentation follows AIGuardian's preference for conceptual-level documentation with visual diagrams and developer-friendly format.