Date: 2025-09-14
Status: Accepted
Deciders: Architecture Team
Context#
The Lilith platform requires a scalable architecture to support multiple AI-powered services including chat, content delivery, speech synthesis, and user management. Key requirements:
- Independent deployment and scaling of components
- Team autonomy for parallel development
- High availability and fault isolation
- Sub-second response times for chat
- Integration with multiple AI providers
Decision#
Adopt a microservices architecture using Fastify as the web framework with a Backend-for-Frontend (BFF) pattern.
Key Components#
- Individual services: auth, content, conversation, ai, media, notification, moderation
- BFF service for client aggregation
- Istio service mesh for inter-service communication
- Event-driven communication for async operations
Rationale#
Why Microservices#
- Enables independent scaling (AI vs. user management)
- Supports technology diversity where needed
- Provides fault isolation
- Matches Kubernetes deployment target
Why Fastify#
- Superior performance vs Express.js
- Built-in schema validation
- Strong TypeScript support
- Good observability hooks
Why BFF#
- Different clients need different data shapes
- Reduces client-server round trips
- Centralizes cross-cutting concerns
Alternatives Considered#
- Monolith: Rejected due to scaling requirements
- Express.js: Rejected for performance reasons
- GraphQL Federation: Rejected due to caching complexity
Consequences#
Positive#
- Better scalability and resilience
- Team autonomy
- Technology flexibility
Negative#
- Distributed system complexity
- Network latency between services
- More complex debugging
Mitigations#
- Comprehensive distributed tracing
- Circuit breakers and retries
- Service mesh for reliability