Home Features Solutions Technology About Training Data Provider Dashboard Training Data Consumer Dashboard Confidential Clean Room Provider Dashboard System Administrator Dashboard API Reference Architecture Guide User Guide Security Guide Compliance Guide Getting Started SDK Documentation Integration Guide Code Examples Deployment Guide Get Started

System Overview

Confidential AI Network is built on a microservices architecture that enables secure, privacy-preserving AI model training through smart contracts and confidential computing.

Microservices Architecture

Built with Node.js microservices that communicate through REST APIs and message queues, ensuring scalability and maintainability.

PostgreSQL Database

Centralized data storage with proper indexing and relationships for users, contracts, datasets, and AI models.

Core Components

Keycloak Identity Provider

Centralized authentication and authorization service that manages user identities, roles, and permissions across the platform.

  • Single Sign-On (SSO)
  • Role-based Access Control
  • Multi-factor Authentication
Contract Service

Manages Ricardian contracts between data providers, consumers, and clean room providers with automated execution.

  • Contract Creation & Management
  • Automated Execution
  • Compliance Validation
SCITT CCF Integration

Microsoft's SCITT CCF provides transparent, verifiable data provenance and audit trails for all data operations.

  • Data Provenance
  • Audit Trails
  • Transparency
Confidential Computing

Enables AI model training on encrypted data without exposing raw information, ensuring mathematical privacy guarantees.

  • Encrypted Processing
  • Privacy Preservation
  • Secure Execution

Authentication & Authorization

Multi-layered security approach with Keycloak managing user identities and role-based access control.

User Registration & Authentication Flow
  1. User Registration: Users register through Keycloak with email verification
  2. Role Assignment: Users are assigned roles (TDP, TDC, CCRP, Admin) based on their needs
  3. Authentication: JWT tokens provide secure access to platform services
  4. Authorization: Role-based permissions control access to different platform features
Role-Based Access Control
Training Data Provider (TDP)
  • Upload and manage datasets
  • Create licensing agreements
  • Monitor usage and revenue
Training Data Consumer (TDC)
  • Browse available datasets
  • Request data access
  • Train AI models securely
Confidential Clean Room Provider (CCRP)
  • Provide secure computing environments
  • Execute training jobs
  • Ensure data isolation
System Administrator
  • Manage platform configuration
  • Monitor system health
  • Handle user support

Contract Management

Ricardian contracts enable automated, legally enforceable agreements between platform participants.

Contract Signing Process
1. Contract Creation

Data providers create contracts specifying dataset terms, pricing, and usage restrictions.

2. Consumer Review

Data consumers review contract terms and accept or negotiate modifications.

3. Digital Signatures

All parties digitally sign the contract using their authenticated identities.

4. Automated Execution

Smart contracts automatically enforce terms and handle payments.

Contract Types
  • Data Licensing Contracts: Define how datasets can be used, for how long, and at what cost
  • Service Level Agreements: Specify CCRP performance and security requirements
  • Compliance Contracts: Ensure regulatory requirements are met for data handling

Data Flow Architecture

Secure data handling from upload through AI model training, with privacy preserved at every step.

Data Processing Pipeline
  1. Data Upload: TDPs upload encrypted datasets with metadata
  2. Contract Agreement: TDCs and TDPs agree on terms through smart contracts
  3. Secure Transfer: Data is transferred to CCRP environments using encrypted channels
  4. Confidential Processing: AI models train on encrypted data without decryption
  5. Result Delivery: Training results are delivered to TDCs without exposing raw data
Privacy Preservation Techniques
Differential Privacy

Adds controlled noise to data to prevent individual identification while preserving statistical properties.

Confidential Computing

Hardware-enforced secure enclaves with attestation and memory encryption for privacy-preserving computation.

Security Architecture

Multi-layered security approach protecting data, communications, and system integrity.

Security Layers
Network Security
  • TLS/SSL encryption for all communications
  • VPN access for secure remote connections
  • Firewall protection and intrusion detection
Data Security
  • End-to-end encryption for data at rest and in transit
  • Key management through secure HSM integration
  • Regular security audits and penetration testing
Application Security
  • Input validation and sanitization
  • SQL injection prevention
  • Cross-site scripting protection
Access Control
  • Multi-factor authentication
  • Role-based access control
  • Session management and timeout

Deployment Architecture

Flexible deployment options supporting both cloud and on-premises environments.

Deployment Options
Cloud Deployment
  • AWS, Azure, GCP, and OCI support
  • Auto-scaling and load balancing
  • Managed database services
  • CDN for global performance
On-Premises
  • Docker container deployment
  • Kubernetes orchestration
  • Local database and storage
  • Custom security policies
Infrastructure Components
  • Load Balancers: Distribute traffic across multiple service instances
  • API Gateways: Manage API access, rate limiting, and authentication
  • Message Queues: Handle asynchronous communication between services
  • Monitoring: Real-time system health and performance monitoring