Pub Finder

Find the optimal pub to meet with your friends in Prague, using public transit optimization.

Try the live demo at https://pub-finder.hermandaniel.com.

How does it work?

We have 1,400+ transit stops in Prague. Given a set of k stops where friends are starting from (e.g. Krymska, Andel, Muzeum), we find the target stop that is closest to everyone. "Closest" can mean:

1. Minimize worst-case -- minimize the maximum travel time from any friend's starting stop.
2. Minimize total -- minimize the sum of all travel times.

The naive approach uses geographic (Haversine) distance between stops. But public transit speeds vary by route, so we scraped ~2.1M stop-pair travel times from DPP to use actual transit minutes as the distance metric.

The search works in stages: first, we select the top 10 stops by geographic distance and top 25 by pre-computed transit time. Then we scrape real-time travel times for the selected date/time and re-rank. The top 15 target stops are returned, each with nearby pubs discovered via the Google Places API (cached for 90 days).

Features

• Session-based -- create a session, share the code, friends join and pick their stops
• Real-time updates -- participant list updates live via Server-Sent Events
• Interactive map -- Leaflet.js map showing stops and recommended pubs
• Pub discovery -- Google Places API integration with rating, price level, and walking directions
• Shareable results -- permanent link to search results for each session
• Round-trip support -- optionally set a different return stop

Quick start

Local development

Requires Python 3.12+.

pip install -e ".[dev]"
cp .env.example .env
# Edit .env and add your GOOGLE_PLACES_API_KEY
python -m backend

Visit http://localhost:3000.

Docker

cp .env.example .env
# Edit .env and add your GOOGLE_PLACES_API_KEY
docker compose up --build

Visit http://localhost:3000.

Environment variables

Variable	Default	Description
GOOGLE_PLACES_API_KEY	(empty)	Required for pub search
DATABASE_PATH	pub_finder.db	SQLite database path
HOST	0.0.0.0	Server bind address
PORT	3000	Server port

Project structure

backend/          FastAPI app, config, DB, optimization, Places API client
routers/          Route handlers (home, session, search)
templates/        Jinja2 templates with HTMX partials
static/           CSS, JS, favicon
data/             Pre-computed transit data (parquet) and stop lists
data_preparation/ CLI tools for scraping and preparing transit data
tests/            Pytest test suite

Testing

pip install -e ".[dev]"
pytest

Data preparation

The data_preparation module provides a CLI for scraping transit times and preparing stop data.

pip install -e ".[data-prep]"
python -m data_preparation --help

Subcommands:

• scrape -- scrape travel times between stop pairs from DPP
• manage -- filter errors and manage scrape results
• prepare -- generate geo data from raw GPS JSON files
• bandit-sim -- run multi-armed bandit simulation for adaptive scraping

Sources

• https://spojeni.dpp.cz/ -- DPP transit journey planner
• https://pid.cz/zastavky-pid/zastavky-v-praze -- PID stop listings
• https://mapa.pid.cz/ -- PID transit map

License

MIT