███████╗██╗██████╗ ██╗    ██╗  ████████╗██╗███╗   ███╗███████╗███████╗
 ██╔════╝██║██╔══██╗██║    ██║  ╚══██╔══╝██║████╗ ████║██╔════╝██╔════╝
 █████╗  ██║██║  ██║██║ █╗ ██║     ██║   ██║██╔████╔██║█████╗  ███████╗
 ██╔══╝  ██║██║  ██║██║███╗██║     ██║   ██║██║╚██╔╝██║██╔══╝  ╚════██║
 ███████╗██║██████╔╝╚███╔███╔╝     ██║   ██║██║ ╚═╝ ██║███████╗███████║
 ╚══════╝╚═╝╚═════╝  ╚══╝╚══╝      ╚═╝   ╚═╝╚═╝     ╚═╝╚══════╝╚══════╝

Live security queue wait times for Dublin Airport Terminal 1 & Terminal 2. The data is already there. This just predicts it.

┌──────────────────────────────────────────────────────────────────────┐
│  Dublin Airport API ──► Scrapy pollers ──► PostgreSQL ──► Flask API  │
│                                                                      │
│  T1/T2 security times   polled every few minutes, stored forever     │
│  Departure board         polled daily, upserted on conflict          │
│  ML predictions     6 models (T1/T2 × 60/120/180 min ahead)           │
│  Monte Carlo sims        Liminal (A/B/D) + Trition (A/B/C/D)          │
│  React dashboard         live tiles, charts, recommendations, PWA    │
└──────────────────────────────────────────────────────────────────────┘

What it is

Real-time dashboard for Dublin Airport (DUB/EIDW) security queue wait times. Shows T1 and T2 times, recommends which terminal to use, shows departure boards, historical trends, and predicted future wait times. Passengers at Dublin Airport can use either terminal's security regardless of which terminal their flight departs from — the app uses that to recommend the shorter queue.

Live at eidwtimes.xyz

Architecture

┌─────────────────┐     ┌─────────────────┐     ┌──────────────┐
│ poll_security.py │────►│                 │     │              │
│ (Cloud Run Job)  │     │   PostgreSQL    │◄────│  Flask API   │
│                  │     │                 │     │  (Cloud Run) │
│ poll_departures  │────►│  6 tables:      │     │              │
│ (Cloud Run Job)  │     │  security_times │     │  /api/*      │
└─────────────────┘     │  departures     │     └──────┬───────┘
                        │  announcements  │     ┌──────┴───────┐
                        │  sec_current    │     │  rate limit  │
                        └─────────────────┘     │  response $  │
                                                └──────────────┘
                                                       │
                                                ┌──────┴───────┐
                                                │  React SPA   │
                                                │  (Vercel)    │
                                                └──────────────┘

Pollers — Scrapy spiders running as Docker containers on Cloud Run Jobs. poll_security.py hits Dublin Airport's security times API. poll_departures.py hits their departure listing API. Both write to Postgres.

Backend — Flask on Cloud Run behind Gunicorn. Serves all /api/* endpoints. Depends on Redis for rate limiting (token bucket, Lua script) and response caching (30s TTL). If Redis is down, all API requests return 503.

Frontend — React 18 + TypeScript + Vite, deployed to Vercel. Uses shadcn/ui, Recharts, Three.js background. PWA-installable.

Bot protection — Fingerprint Pro device fingerprint + reCAPTCHA Enterprise → JWT ("bounce token"). All API calls require this JWT. Requests also go through a "Datagram" layer that hashes API routes and signs them with HMAC-SHA512.

Prediction models

Two engines, user picks via cookie (forecast_model):

Liminal — Monte Carlo simulations. No ML. Methods A (multi-path random walk), B (projected trend), D (departure-aware sim).

Trition — ML. 6 models: T1 and T2, each predicting 60/120/180 minutes ahead. Trained on Modal (A100 GPU), logged to W&B, published to Hugging Face Hub (xgboost-sm). Downloaded at Docker build time, not runtime. Features: 24h of lagged 5-min security bands, departure counts in sliding windows, sine/cosine time encoding.

Data sources

Source	What	How
api.dublinairport.com/dap/get-security-times	T1/T2 queue minutes	Scrapy spider, every few min
api.dublinairport.com/dap/flight-listing/departures	Departure board	Scrapy spider, 2-day lookahead
Google Maps Routes API	Drive times from 150+ origins	Periodic polling
Hugging Face Hub	XGBoost model weights	Downloaded at build

Running locally

# frontend
cp .env.example .env.local
pnpm install
pnpm dev

# backend
cd backend
pip install -r requirements.txt
# needs DATABASE_URL, REDIS_URL, BOUNCE_TOKEN_SECRET, DATAGRAM_SIGNING_KEY
python -m flask run

Deploying

deploy.bat [PROJECT_ID] [REGION]

This builds 3 Docker images (security poller, departure poller, backend), pushes to GCR, runs Terraform, and inits the DB schema.

Environment variables

Frontend (.env.local):

Var	What
VITE_API_BASE_URL	Backend URL (e.g. https://datagram.eidwtimes.xyz)
VITE_GA_TRACKING_ID	Google Analytics
VITE_POSTHOG_KEY	PostHog analytics
VITE_RECAPTCHA_SITE_KEY	reCAPTCHA Enterprise site key
VITE_FINGERPRINT_API_KEY	Fingerprint Pro
DATAGRAM_SIGNING_KEY	HMAC key for Datagram routing

Backend (backend/.env):

Var	What
DATABASE_URL	Postgres connection string
REDIS_URL	Redis connection string
HF_TOKEN	Hugging Face token (model download)
BOUNCE_TOKEN_SECRET	JWT signing secret
DATAGRAM_SIGNING_KEY	HMAC key for Datagram verification
RECAPTCHA_SITE_KEY	reCAPTCHA Enterprise
GCP_PROJECT_ID	GCP project for reCAPTCHA
PORT	Gunicorn bind port

Database

Schema in sql/schema.sql. Run with:

python setup-db.py <connection_string>

All tables use IF NOT EXISTS — safe to re-run on every deploy.

License

See LICENSE.