Relief SF - A restroom finder built using codex

Finding the best and easiest way to heed nature's call without the stress of guessing where to go. Focus on finding the best restroom. Built with codex, powered by the community, scaled with GPT 5.6.

Comment

Inspiration

I've personally struggled with finding restrooms in public areas in large cities. Google maps never quite cut it because its mostly photos of food and menus. As an avid runner I've had to build a mental map of SF's restrooms. Lets be real, not all are created equal. I'm also a Google Local Guides power user and contribute quite a bit. I thought it'd be awesome to build a community driven real-time map of usable restrooms in SF and eventually scale to other cities like NYC.

What it does

Relief SF allows anyone with a browser to search and find a restroom and see key pieces about that restroom. For example, is it a coffee shop restroom that expects a purchase or a free restroom you can pop into on a Sunday night.

When the need is urgent, I’m Feeling Lucky uses the person’s location to find the physically closest option.

Anyone can also contribute without creating an account. A contributor can update an existing restroom or suggest a missing business through Mapbox place discovery, then add access information, a cleanliness rating, notes, and a restroom-only photo.

Submissions remain private while GPT-5.6 and, when needed, a human operator review the evidence.

How I built it

-As a former 0->1 builder I used Codex as my building partner. I gave codex the vision, the plan, the scope, requirements, and user flows. Built over a weekend focusing on low user friction (i.e. not an iphone app) and easy contribution (no auth layer) but with scaled guardrails in mind (human review operator dashboard).

-Vercel hosted, supabase db, opensource map seeds (i.e. DataSF and OpenStreetMap) and enrichment search with Mapbox.

Challenges we ran into

  • Seeding the initial map: We intentionally avoided crawling Yelp or Google. Codex helped identify DataSF’s public-restroom dataset as a permitted starting point.
  • Limited restroom photography: We accepted that the initial map would not have many photos. Instead, we made personal contributions, suggestions, and photo uploads easy, with no account required.
  • Guardrails: Manually reviewing thousands of restroom leads would not scale. We built an operator workspace where GPT-5.6 reviews evidence and routes community submissions and source-attributed research leads.
  • Deployment: Testing the complete workflow required repeatedly resetting Supabase data and exercising submissions, GPT review, operator decisions, photo publication, and map updates.
  • Scope: We could have attempted nationwide coverage, but chose to make the San Francisco experience useful first.

Accomplishments that we're proud of

  • Real user photos are part of the map

  • Codex helped research and build the system. GPT-5.6 operates inside the finished product.

  • GPT image review allows for gated responsible approval of images

  • No bananas were allowed on to the site as photos

  • Something I've started to use since

What we learned

  • Not all data is created equal. Publicly available data lacks important and structured types. Find a way to leverage community to enrich.

  • You dont need an iphone app to build a cool map application

What's next for Relief SF - A restroom finder built using codex

-Scaling to NYC. We've built the ingestion pipeline for 5.6! And NYC also has an awesome public database of recrods.

Built With

Share this project:

Updates