A universal Slovak train departure board
Pick a station · live-counting departures · Cloudflare Worker + D1 · PWA
Architecture, deployment, and GTFS data refresh — published as a documentation site with sidebar navigation (EN / SK). This README is the summary.
🚆 Live: vlaky.ensecnet.net
This board is small on purpose. It's a working demonstration that a genuinely useful public service can be built from open data, a free cloud tier, and modern AI-assisted development — without a data centre, a budget, or a team.
- Open state data, used properly. The timetable comes from the official GTFS feed the state publishes on data.slovensko.sk. Public data is there to be built on — this is what that looks like in practice: take the feed, transform it, give people something they can actually use.
- Minimal infrastructure, near-zero cost. One Cloudflare Worker, one D1 database, a 45-second edge cache. No origin server, nothing running 24/7 to pay for — it serves hundreds to thousands of people a day inside the free tier. The whole architecture fits on a napkin.
- Built fast with AI vibe coding. From idea to a deployed PWA with search, GPS, trip routes, stats, and an ops dashboard — built quickly with AI-assisted development. The barrier between "I have an idea" and "it's live on a domain" has collapsed.
The point: when the will is there, a path opens — even for things that look like they need a big system, a budget, and months of work. Open data + a free cloud edge + AI as a force multiplier turns a weekend into a real, public, installable app. That's the demonstration.
A departure board for Slovak train stations: choose a station and see three just-left departures (dimmed) plus the next fifteen — each with destination, train number, and a live countdown. Plan times come from GTFS (ZSSK / ŽSR); no live delays.
Built as a single Cloudflare Worker with a D1 (SQLite) database and a 45-second edge cache — no origin server, runs inside the free tier.
- Station search with an accent-normalising autocomplete
- Favourites kept in
localStorageon the device - Per-station URL (
/{station}) — bookmark or pin to the home screen - 📍 GPS nearest station — coordinates are used and discarded, never stored
- PWA — installable, app-like
- Live countdown — "in X′ / now"
- 45-second edge cache — absorbs traffic spikes, keeps D1 idle
- Anonymous aggregate stats — day · station · hour · region; no IPs, no individuals
/opsdashboard — operational view behind login
GPS coordinates for the nearest-station lookup are used in the request and thrown away — never stored. Statistics are aggregate only (day, station, hour, region from the Cloudflare edge); no IP addresses, no individual tracking.
Tap any departure and the board opens that train's full route — every station it
calls at (/vlak/{number}), with arrival and departure times. Stations already
passed are dimmed; stations still ahead are bright, so the current position reads
at a glance.
Every station has its own URL — append the station to the address and the board opens straight on it:
https://vlaky.ensecnet.net/bratislava-zelezna-studienka
Bookmark it, or use the browser's "Add to Home Screen" to drop an icon that opens your station in one tap — effectively a per-station app shortcut.
The whole thing is built as a PWA (Progressive Web App), so it installs straight onto your phone from the browser — no App Store, no Google Play. Open the site, choose "Add to Home Screen" / "Install", and it lands as an icon and runs full-screen like a native app. Favourites and your pinned station travel with it.
cd ~/vlaky && npm install
npx wrangler d1 create vlaky # paste database_id into wrangler.toml
npx wrangler d1 execute vlaky --remote --file=./migrations/0001_init.sql
node build-d1.js gtfs.zip # build timetable from GTFS
npx wrangler d1 execute vlaky --remote --file=./data/seed.sql
npx wrangler deploy # binds vlaky.ensecnet.net (custom_domain)Full walk-through — schema, seeding, /ops login, custom domain — is in the
documentation.
vlaky/
├─ src/index.js # Worker: board, search, geo, stats, /ops, PWA
├─ migrations/0001_init.sql # D1 schema
├─ build-d1.js # GTFS zip → data/seed.sql + data/stations.json
├─ data/seed_sample.sql # small sample for a pre-GTFS test
├─ update.ps1 # one-shot GTFS refresh (Windows)
├─ wrangler.toml
└─ package.json
Served from the docs/ folder via GitHub Pages (EN primary, SK switch):
| Page | What's in it |
|---|---|
| Overview | Features, privacy model |
| Architecture | Worker + D1, edge cache, endpoints |
| Deployment | From npm install to bound domain |
| GTFS data refresh | Rebuilding the timetable |
Free tier: Worker 100k req/day, D1 5 GB + 5M reads/day. With the 45-second edge cache that holds hundreds to thousands of daily users for free. Times are GTFS plan, not realtime — a realtime board would need a GTFS-RT feed, which Slovakia doesn't publish openly yet (there's a hook for it in the Worker).
Set the hero banner as the repository Social preview in Settings → General → Social preview so it shows when the link is shared (LinkedIn, Slack, Discord).
GTFS data © ZSSK / ŽSR, from data.slovensko.sk. Code under this repository's license.