Cloud Infrastructure and Architecture

Architecture overview

AWorld implements a fully serverless cloud-native architecture on Amazon Web Services (AWS), designed to guarantee unlimited scalability and high availability. The architecture is organized in functional layers that separate responsibilities and facilitate platform evolution:

Account & user layer: identity management, permissions, authentication (AWS Cognito), and server-to-server integration

Gamification layer: core engine for engagement mechanics (missions, levels, leaderboards, badges, points)

Catalog layer: distribution and organized management of training content

The multi-tenant architecture guarantees rigorous data isolation between clients, with each workspace operating in complete logical independence while sharing the underlying physical infrastructure for operational efficiency.

Technology stack

The platform is based on AWS services, which guarantee high standards of security, reliability, and reduced infrastructure maintenance requirements.

Component	AWS Service	Function
Compute	AWS Lambda	Serverless execution of API functions
API management	Amazon API Gateway	Routing, throttling, request authentication
Primary database	Amazon DynamoDB	NoSQL database with tenant-scoped composite keys
Dedicated database	Cloudflare D1	Dedicated database for indexing and specific application functions
Caching/Leaderboard	AWS MemoryDB for Redis	In-memory system for real-time leaderboards
Media storage	Cloudflare R2	Media storage with automatic caching and redundant CloudFront
DNS/routing	AWS Route 53	Multi-region geographic load balancing
Authentication	AWS Cognito	Identity management and user pools
Authorization	AWS Verified Permissions	Policy-based fine-grained authorization
Feature flags	AWS AppConfig	Feature flags with safe deployment strategies
Event bus	Amazon EventBridge	Multi-region event-driven architecture
Configuration	AWS SSM Parameter Store	Cross-stage configuration management
Monitoring	AWS CloudWatch	Metrics, logs, and anomaly detection
Security	AWS WAF	Web application firewall

Infrastructure as Code

The entire infrastructure is defined as code (Infrastructure as Code) using SST (Serverless Stack) version 3 with Pulumi backend, guaranteeing:

Declarative definition: Infrastructure written in TypeScript for type safety and compile-time validation

Automated deployment: Reproducible and versioned deployment process

Native multi-region: Declarative configuration for simultaneous deployment across multiple regions

State management: Pulumi engine for infrastructure state management and drift detection

Version control: All infrastructure changes go through code review and CI/CD

This approach eliminates error-prone manual configurations and guarantees consistency between development, staging, and production environments.

Database and data persistence

Amazon DynamoDB

DynamoDB is the primary database for all transactional and operational data on the platform:

Model: NoSQL with composite primary keys that include workspaceId for tenant isolation

Scaling: on-demand capacity mode with automatic scaling based on load

Performance: single-digit millisecond latency for read/write operations

Partitioning: automatic for load distribution

Replication: global tables for real-time multi-region synchronization

Cloudflare D1

Dedicated database alongside DynamoDB to fulfill specific application functions:

Usage: indexing and complex queries

Integration: alongside primary database to optimize specific operations

Multi-tenant model with data isolation

AWorld implements a multi-tenant architecture with rigorous logical isolation that guarantees complete data separation between different clients.

Hierarchical structure

Platform (AWorld)
└── Account (client/tenant)
    └── Workspace (environment: production, staging, dev)
        └── User (end user with specific roles)

Isolation mechanisms

At database level (DynamoDB):

Tenant-scoped keys: each record includes workspaceId as part of the primary key

Automatic query filtering: queries are automatically filtered by workspace

Row-level isolation: physical impossibility of accessing data from different workspaces in the same query

At API Gateway level:

AWS Lambda Authorizer validates every request

The JWT token contains the workspaceId claim critical for isolation

Cross-workspace access blocked before business logic execution

At Cognito level:

Shared user pool with workspace-isolated credentials

Dedicated Cognito app client per workspace

Workspace-scoped JWT tokens

⚠️ Critical security note: The workspaceId in the JWT token is the fundamental mechanism for multi-tenant isolation. Each API request validates that the workspaceId in the token matches the workspace of the requested resources, preventing cross-tenant access at the API Gateway level before the request reaches the backend.

Multi-region distribution

To guarantee high availability and disaster recovery, the AWorld infrastructure is distributed across multiple AWS regions in active-active configuration.

Operating regions

Primary region: eu-west-1 (Ireland) - Europe production

Backup region: eu-north-1 (Stockholm) - disaster recovery

Data residency: 100% of data maintained in EU data centers for GDPR compliance

Geographic load balancing

AWS Route 53 manages intelligent traffic routing:

Geolocation routing: users are directed to the geographically closest region

Continuous health checks: continuous monitoring of regional status

Automatic failover: in case of regional failure, traffic is automatically diverted to the backup region

Latency-based routing: automatic optimization for lower latency

Advantages of serverless architecture

Adopting a serverless architecture offers significant benefits in operational and reliability terms.

Automatic scalability

AWS Lambda scales automatically based on request volume

No manual provisioning of servers or capacity

Transparent handling of unpredictable traffic spikes (e.g., corporate engagement campaigns)

Additional instances activated in milliseconds

Reduced operational management

Patching and updates managed automatically

Zero downtime for infrastructure maintenance

Technical team focus on feature development instead of server maintenance

Intrinsic resilience

Integrated fault tolerance

Automatic distribution across multiple availability zones

Significant reduction in single point of failure risk

Cloud Infrastructure and Architecture

Architecture overview#

Technology stack#

Infrastructure as Code#

Database and data persistence#

Amazon DynamoDB#

Cloudflare D1#

Multi-tenant model with data isolation#

Hierarchical structure#

Isolation mechanisms#

Multi-region distribution#

Operating regions#

Geographic load balancing#

Advantages of serverless architecture#

Automatic scalability#

Reduced operational management#

Intrinsic resilience#