Day & Night DC — Tyler Zanona

The city in two layers

I love coffee. I love cocktails. I love DC. By day Washington DC is a coffee city — roasters and neighborhood shops on every other corner. By night, some of its best bars hide behind unmarked doors. Day & Night DC is a hyperlocal guide to both: flip between Day and Night, and the map shows you where to go right now.

The map is half the pleasure: a dark, metro-style view of the city drawn around Washington's actual Metro lines. Tap any station and you get live train arrivals straight from WMATA — so you know exactly how long you have for one more round.

Underneath runs a data pipeline I built: an AI extraction system reads local-business websites and pulls structured information — hours, locations, contact details, and more. From there, I verify things by hand and curate the absolute best and brightest of the coffee shops and speakeasies in the dataset.

Field research

Instant-film photo of Tyler in a dim bar booth, drink in hand — Field research — shot on instant film

On the map

The Day & Night DC night map: Washington drawn around its Metro lines, speakeasy spots marked in violet — The night map

Live train arrivals at Metro Center, straight from WMATA — Live Metro arrivals via WMATA

Only the best make the map

Behind the main app sits an admin app I built to train the discovery model. It surfaces speakeasy and coffee-shop candidates from open venue data (2,304 and counting), and I judge each one: keep, review later, or reject — tagging the keepers with what makes them belong.

The Day & Night DC admin trainer: a candidate tagged true-speakeasy, perfect-vibe, and hidden, with Keep, Review later, and Reject buttons — The review room — keep, later, or reject

A speakeasy candidate card in the admin trainer: a Dupont Circle cocktail lounge with category and license tags — Candidate 1 of 2,304 — judged by hand

How it's built

PWA today, iOS on the roadmap — live in the browser now, headed for the App Store (eventually).
Metro-style map — dark cartography drawn around the city's real transit lines.
Live WMATA data — tap a station for real-time train arrivals.
A model with a human teacher — an admin trainer surfaces candidates from open venue data (Overture Maps, Foursquare) and learns from every keep and reject.
TypeScript on Cloudflare Workers — a serverless pipeline with a D1 (SQLite) database of geocoded, human-verified spots.

What's next

Currently building: the compass — point your phone and it points you to the nearest coffee shop or speakeasy, depending on the hour.