Vorgo

Here’s the same Devpost writeup, clean and professional with no emojis:

Inspiration

Modern apps do not fail in obvious ways. They fail under load, during traffic spikes, or when a single component crashes. We wanted to build a system that does not just work for one user, but continues working under stress and failure.

This project was inspired by real production systems where uptime, scalability, and observability are critical. We focused on answering one question: what happens when everything goes wrong?

How we built it

We built a distributed URL shortener designed for resilience and scale.

Backend: Flask API for URL shortening and redirects
Database: PostgreSQL for persistent storage
Cache: Redis to reduce repeated database queries
Scaling: Multiple containerized app instances using Docker
Load balancing: Nginx distributes traffic across instances
Load testing: k6 simulates hundreds of concurrent users
Observability: Prometheus and Grafana for metrics and dashboards
Alerts: Automated alerts sent via Discord when failures occur

The architecture follows a horizontally scalable model where traffic is distributed across multiple services:

Client → Nginx → App Instances → Redis / PostgreSQL

What we learned

Horizontal scaling is more effective than increasing server power
Caching significantly reduces latency and database load
Systems must be designed to fail gracefully, not perfectly
Observability is essential for understanding system behavior
Reliability comes from testing failure scenarios, not avoiding them

Challenges we ran into

Container crashes: ensuring automatic recovery without downtime
Load balancing: correctly routing traffic across multiple instances
Caching strategy: avoiding stale data while improving performance
High concurrency: handling hundreds of simultaneous requests without errors
Debugging under load: identifying bottlenecks between CPU, database, and network

One key bottleneck was repeated database reads. By introducing Redis caching, we reduced unnecessary queries and improved response time under heavy load.