PE hackathon 2026 - Production grade url shortener

API Routes and url testing

Inspiration

We wanted to build something deceptively simple — a URL shortener — and treat it like a real production system. The challenge wasn't the feature, it was the engineering: how do you make it reliable, fast, and scalable under real load?

What it does

URL Shortener — a production-grade link shortening service with a full REST API.

Users can submit any long URL and receive a short code (e.g. http://206.189.59.175:8000/aB3xZ9). Accessing the short link redirects instantly to the original URL. The system tracks every redirect as a click event, stores user and URL records, and exposes full CRUD endpoints for managing users, links, and events.

Core features:

POST /shorten — generate a short URL with an auto-generated or custom short code
GET / — redirect to the original URL (Redis-cached for speed)
POST /urls — create a URL with title, user ownership, and metadata
GET /users, POST /users — manage user accounts with email validation
GET /events — query click events filtered by URL, user, or event type
Automatic click tracking on every redirect
Graceful Redis fallback — if cache is unavailable, falls back to the database
Full input validation — rejects malformed requests, invalid emails, bad URL schemes
GET /health — health check endpoint for monitoring

How we built it

We started with a minimal Flask app and iterated through three engineering tiers for both reliability and scalability:

Reliability track: Added a /health endpoint, built a GitHub Actions CI pipeline, wrote 100+ tests with 95% coverage, implemented structured error handling, and configured restart: always for chaos resilience.

Scalability track: Ran k6 load tests at 50, 200, and 500 concurrent users. Fixed bottlenecks at each stage — moving from a single Flask process to Gunicorn workers, adding Nginx load balancing across two app containers, adding connection pooling, and finally Redis caching for redirect lookups.

Deployment: Manually provisioned a DigitalOcean Droplet (CentOS 9), installed Docker, and ran the full stack via Docker Compose — live at http://206.189.59.175:8000.

Challenges we ran into

Nginx crash-looping because it couldn't resolve app container DNS before they fully joined the Docker network
- gevent async workers breaking psycopg2 due to C extension incompatibility — had to revert to sync workers
- PostgreSQL connection exhaustion under 200 VUs before we added connection pooling
- CI test failures caused by patching the wrong class (PostgresqlDatabase vs PooledPostgresqlDatabase)

Accomplishments that we're proud of

What we learned

We passed majority tests, that's no small feat

How horizontal scaling with Nginx and multiple app instances actually works at the infrastructure level
The difference between connection pooling and opening raw DB connections — and why it matters under load
How Redis caching dramatically reduces database pressure on hot read paths
How to provision and deploy a production server on DigitalOcean from scratch — no managed platforms, just a Linux VM and Docker
How CI/CD gates (coverage thresholds, test requirements) enforce quality before code reaches production