A live operations dashboard for a railway station. An admin selects a state, city, and station, then sees real‑time platform occupancy, the train board, and smart platform suggestions generated by a rule‑based allocation engine.
Live demo: https://switch-yard.vercel.app API: https://switchyard-scpr.onrender.com · API docs: https://switchyard-scpr.onrender.com/docs
Note: the backend runs on a free tier that sleeps after ~15 minutes idle, so the first request after a quiet period can take 30–50 seconds to wake up. It's fast after that.
- Cascading selectors — pick State → City → Station, all loaded from the database.
- Four live stat cards — total platforms, free, occupied, and trains arriving in the next hour. Each card is clickable for a breakdown.
- Platform status board — a colour‑coded tile per platform (green = free, amber = arriving, red = occupied).
- Train board — one table of every train (arrival, departure, expected time, platform, status) with a merged column showing the engine's platform suggestion, and the reason on hover.
- Platform allocation engine — when a train is delayed and clashes with another on its platform, the engine proposes the cheapest fix (relocate the delayed train, relocate the other train, or hold one) measured in added‑delay minutes.
- Dark mode, a live clock, and auto‑refresh every 20 seconds.
| Layer | Tech |
|---|---|
| Frontend | React, Vite, Tailwind CSS v4 |
| Backend | FastAPI, SQLAlchemy, Pydantic |
| Database | PostgreSQL |
| Hosting | Vercel (frontend), Render (backend + managed Postgres) |
SwitchYard/
├── backend/ # FastAPI + SQLAlchemy API
│ ├── app/
│ │ ├── api/routes/dashboard.py # HTTP endpoints
│ │ ├── config/settings.py # env-driven config
│ │ ├── db/ # models, session, crud, seed + table scripts
│ │ ├── logic/allocator.py # platform-allocation engine
│ │ └── main.py # app entry point + CORS
│ ├── Dockerfile
│ ├── requirements.txt
│ └── .env.example
├── frontend/ # React + Vite + Tailwind dashboard
│ └── src/
│ ├── App.jsx # state, data loading, layout
│ ├── api.js # backend URL + fetch helpers
│ └── components/ # Header, Selectors, StatCards, PlatformBoard, TrainBoard, DetailPanel
├── docker-compose.yml # local Postgres for development
└── README.md
Prerequisites: Python 3.11+, Node 20+, and Docker (for the local database).
From the repository root:
docker compose up -d dbThis runs PostgreSQL on localhost:5432 (user postgres, password postgres, database switchyard).
cd backend
python -m venv venv
# Windows: venv\Scripts\activate macOS/Linux: source venv/bin/activate
pip install -r requirements.txt
# create your .env from the template
cp .env.example .env # Windows: Copy-Item .env.example .env
# create the tables and load sample data
python -m app.db.reset_table
python -m app.db.seed_data
# run the API
uvicorn app.main:app --reloadThe API is now on http://localhost:8000 — check http://localhost:8000/docs.
cd frontend
npm install
npm run devOpen http://localhost:5173. In development it calls http://localhost:8000 by default, so no extra config is needed.
The sample data is anchored to the moment you seed it and stays realistic for about half a day. If the board ever shows everything as "departed", just re-run
python -m app.db.seed_datato refresh it.
Backend
| Variable | Purpose | Example |
|---|---|---|
DATABASE_URL |
PostgreSQL connection string | postgresql://postgres:postgres@localhost:5432/switchyard |
FRONTEND_ORIGIN |
Allowed CORS origin(s), comma‑separated | https://switch-yard.vercel.app |
Frontend
| Variable | Purpose | Example |
|---|---|---|
VITE_API_URL |
Base URL of the backend API (no trailing /api) |
https://switchyard-scpr.onrender.com |
backend/app/logic/allocator.py is pure Python — no ML, every decision traceable to a
timestamp comparison. For each delayed train that clashes on its platform it:
- builds an occupancy map of every platform's bookings,
- detects conflicts using a half‑open interval overlap test,
- assembles candidate fixes — relocate the delayed train to the nearest free platform, relocate the on‑time train it clashes with, or hold the on‑time train so they share the platform in turn,
- prices each option in added‑delay minutes and keeps the cheapest,
then writes a suggestion (with a human‑readable reason) that the dashboard shows in the train board.
The frontend is a static Vite build hosted on Vercel (root directory frontend,
with VITE_API_URL pointed at the backend). The backend is a Docker service on Render
(root directory backend, using backend/Dockerfile) connected to a Render managed
PostgreSQL instance via DATABASE_URL, with FRONTEND_ORIGIN set to the Vercel URL so
CORS only permits the real site. Both platforms redeploy automatically on every push to
main.
This project is for learning and demonstration purposes.