Database Architecture

The platform uses both PostgreSQL and MongoDB, chosen based on each application's data model requirements.

Database Dependency Map

Technology Selection

Database	Applications	Rationale
PostgreSQL (Neon + pgvector)	Bookmarked	Semantic search with vector embeddings, relational bookmark/tag structure
PostgreSQL (in-cluster)	Bookmarked, Code Talk, TenantFlow, Pop!_Portfolio	Relational data with strong consistency requirements
MongoDB	educationELLy, educationELLy GraphQL, IntervalAI	Flexible document schemas for educational content and flashcards
MongoDB	Microservices, Vertex Platform	Application data in dedicated namespaces
Redis	Code Talk	Session caching and real-time pub/sub for chat functionality

PostgreSQL

Bookmarked

Bookmarked uses two PostgreSQL instances:

• Neon (external): Serverless PostgreSQL with pgvector for semantic search — bookmark descriptions are embedded as vectors for similarity search, plus tsvector for keyword-based full-text search
• In-cluster: postgresql-bookmarked StatefulSet for the primary relational data (bookmarks, tags, users) — backed up daily via pg_dump CronJob

Code Talk

Code Talk uses PostgreSQL for:

• User accounts and authentication
• Chat rooms and message history
• Code snippet storage with syntax highlighting metadata

TenantFlow

TenantFlow uses PostgreSQL with tenant isolation:

• Row-level security for multi-tenant data isolation
• Tenant-specific configuration and settings

MongoDB

Resource Configuration

MongoDB StatefulSets run without CPU limits to avoid CFS throttling, which can cause crash loops under bursty write workloads. Memory limits are still enforced. Health probe timeouts are set wider than defaults (10-15s) to account for MongoDB's startup time and journal recovery.

educationELLy / educationELLy GraphQL

Both applications share a MongoDB instance for:

• Lesson plans with nested exercise structures
• Student progress tracking with embedded documents
• Flexible content schemas that vary by lesson type

IntervalAI

IntervalAI uses MongoDB for:

• Flashcard decks with spaced repetition metadata
• User learning history and interval schedules
• AI-generated card content storage

Namespace-Isolated Databases

Microservices Namespace

The microservices namespace runs its own MongoDB instance (mongodb) and Redis cluster (redis-replicas-{0,1,2}). These are isolated from the default namespace databases via Kubernetes namespace boundaries.

Vertex Platform Namespace

The vertex-platform namespace runs MongoDB, Redis, and InfluxDB:

• MongoDB: Application data storage
• Redis: Caching and session management
• InfluxDB: Time-series data for platform metrics

Pop!_Portfolio

Pop!_Portfolio uses an in-cluster PostgreSQL instance for application data storage.