Building a Real-Time Analytics Platform for Citrus Systems

Project Background

Citrus Systems, a rapidly growing SaaS company, was struggling with their legacy analytics infrastructure. Their existing solution was built on a monolithic architecture that couldn't scale with their growth.

Key Challenges

1. Performance degradation - Query times had grown to 30+ seconds
2. Scaling limitations - The system couldn't handle traffic spikes
3. High operational costs - Running on oversized dedicated servers
4. Limited flexibility - Adding new metrics required weeks of development

Technical Approach

We designed a modern, event-driven architecture using the JAMstack philosophy extended to backend services.

Architecture Overview

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   Events    │────▶│   Workers   │────▶│   Storage   │
│   (Edge)    │     │  (Process)  │     │  (Query)    │
└─────────────┘     └─────────────┘     └─────────────┘

Phase 1: Event Ingestion

We implemented CloudFlare Workers at the edge to capture and validate events before they hit our core infrastructure. This approach:

• Reduced latency to <10ms for event capture
• Filtered invalid events before storage
• Provided geographic distribution automatically

Phase 2: Processing Pipeline

Events flow through a lightweight processing pipeline built on Node.js:

• Real-time aggregation for dashboards
• Batch processing for historical analysis
• Anomaly detection for alerting

Phase 3: Query Layer

We built a query layer optimized for analytics workloads:

• PostgreSQL with TimescaleDB for time-series data
• Redis for caching frequently accessed aggregates
• Pre-computed rollups for common queries

Implementation Details

Event Schema Design

We designed a flexible event schema that balances performance with extensibility:

• Core fields indexed for fast filtering
• JSONB columns for custom properties
• Partitioning by time for efficient pruning

Caching Strategy

Our multi-layer caching strategy ensures consistent sub-100ms responses:

1. Edge cache - Static aggregates at CDN level
2. Application cache - Redis for dynamic queries
3. Database cache - PostgreSQL query plan caching

Monitoring & Observability

We implemented comprehensive monitoring:

• Real-time dashboards for system health
• Automated alerting for anomalies
• Detailed logging for debugging

Results & Impact

The new platform exceeded all performance targets:

Metric	Before	After	Improvement
Query Time	30s	<100ms	300x faster
Daily Capacity	500K	10M+	20x increase
Monthly Cost	$15,000	$6,000	60% reduction
Uptime	99.5%	99.99%	10x fewer incidents

Lessons Learned

1. Start with the query patterns - Understanding how data would be queried informed our entire architecture
2. Edge processing is powerful - Moving validation to the edge dramatically improved reliability
3. Pre-compute aggressively - Most analytics queries follow predictable patterns

Technologies Used

• Frontend: Eleventy, TailwindCSS, Alpine.js
• Backend: Node.js, CloudFlare Workers
• Database: PostgreSQL + TimescaleDB, Redis
• Infrastructure: CloudFlare, Docker, GitHub Actions