From touring to building GTFS feeds for rural bus agencies

An intervention

I know how to build websites, I can fix this. I threw something together (this was pre-llms) and emailed it to them: https://maxtkc.github.io/wests-bus-sample/

For 1 year, I received no response. The following summer, my parents took the bus from Canada, and I sent them on a special mission to figure out who was in charge.

Subject: [Attention Julie!] Offering Website Assistance

Hi Julie,

My name is Max Katz-Christy, and I am one of your happy passengers and I would like to provide you with a website update. My parents recently used your bus and told me that you are who I should bring this to. I am a software developer, and I’d like to volunteer to help you use make your website:

• easier to search for
• more accessible
• in a more modern format

Here is a sample website I have created that I could easily provide to you: https://maxtkc.github.io/wests-bus-sample/

I also may be able to help you comply with the FTA’s somewhat recent requirement that all transportation agencies provide data to them: https://www.federalregister.gov/documents/2022/07/07/2022-14502/national-transit-database-reporting-changes-and-clarifications

In short, I am a software developer who loves the connection that your bus makes and I have built a website presenting the exact same information that you have on your current website but in a more readable fashion.

About myself, I am 23 years old and an undergraduate student at MIT. I love traveling with my bicycle and on buses and trains. I race sailboats, mountain bikes, and play ice hockey.

I sent you an email about a year ago with this, but I think it may have gotten lost. I’d love to make this happen. There are absolutely no strings attached, I am doing this because I want more people to be able to ride your bus.

Best,

Max

I then had the pleasure of chatting with Julie and Jolyne on the phone, who explained that due to very unfortunate circumstances they had completely lost access to updating their website. Though I was disappointed that they were not interested in my updated website (“we like the red text”), I was happy to help solve their more fundamental issues.

They quickly trusted me so much they sent me all of the passwords they knew about. They were paying for some strange ancient and expensive hosting service, so I moved them to GitHub pages and pointed their domain to it. I made a GitHub organization so they have the same access to it that I have and would not need to email me passwords or share mine. They wanted a banner on the site where they can post updates, so there is a file that they can update, triggering a build that places the text in a yellow banner at the top of the page.

Next, I boosted their SEO to the mooooon! They are the first result for “west bus maine”. Booya. This is done through registering them with Google, proving ownership, and then improving the performance of the page as well as other web “best practices” that they give you brownie points for.

Mission Deux: an email from MaineDOT

We are in need of help with this email and attachment. We have know idea what most of it means or how to do. Would you be able to look it over and let us know if this is something you are able to do for us?

Twist my arm. So, what were they asking for? A GTFS feed.

And what is a GTFS feed, you may ask? It’s a transit agencies timetable, written for computers. It’s a big file describing every scheduled trip on every route that an agency will take. Even my 7:23 red line train from Porter? Yup. The minute it is supposed to depart is in there.

Well, who is using this big stupid file? Why’s it so damn important? While I do like me a good GTFS file for breakfast, it’s also used by the state of Maine for their record keeping, and… you! Do you use Google Maps or Apple Maps (yikes) and navigate with public transit directions? You are using GTFS. Google and Apple each have lists of urls for different agencies GTFS files (confusingly called “feeds”), and every so often they go through every file and download it, providing updated schedules to travelers. This is how you can navigate from Boston to the border of Canada you see the bespoke agencies like west bus: https://maps.app.goo.gl/RnuhJyURxH4MM9H98

For West Bus, I wrote this crappy script to generate their feed: https://github.com/west-bus-service/single-page-site/blob/main/scripts/gen_gtfs.py

It was before AI, so there are no comments in the code, sorry. However it works. A GTFS feed (just know it just means file) is simply a zip file. When you expand it, you’ll see a bunch of .txt (text) files. If you look at any of these, you’ll notice that every single file is really just a csv, which can be loaded in Excel! If you want to dig around deeper, you can check out “the bible” of GTFS: https://gtfs.org/documentation/schedule/reference/ showing exactly how every file should be formatted. You can also download the MBTA’s GTFS from their website and add lots of service along your favorite route exactly to your needs and email them the file as a suggestion: https://www.mbta.com/developers/gtfs

You might notice how the first half of the route to Canada is those ugly straight lines. Concord coach does not in fact operate flying buses that go directly from south station to Maine. They just forgot to include the optional yet coveted shapes.txt file. I sent them a feed too, but they have been ghosting me.

Now, once you write the file, all you’ve got to do is submit it to Google and then, voila, right?

Nope.

Mission Tree: Google QA

It appears that some stops defined in the feed do not match their actual physical locations.

-Google

Yeah. Well, if I’m being honest, I had no clue exactly where they stopped. The website refers to some quite… Maine places: “Swamp Yankee BBQ” (permanently closed) and “Elmer’s Discount”.

You start off by submitting a Google form, with your interest in the program. Next, you need to verify that you are the owner or approved by the owner to submit the schedule. Then, you go back in forth for many months where they take a long time to give you feedback, and then you take a while to respond because you kinda forgot about it, and it goes back and forth until you have a 50+ long email thread with attachments bogging down your Google storage to the point that every month you are deleting various other emails so you can get a few bytes below the 15 GB maximum and evade the impending doom of the “you will not be able to send or receive emails in 3 days if you do not get your shit together” emails.

The process with Google starting in December 2024 managed to take less than one year. In September 2025, West Bus went live on Google!

I should note that it also works on the other mapping services. However, they, including Apple, do not have an independent submission process. They use these lists that you can sumbit the url to, and honestly I have no clue if they do any vetting at all.

A Regulah: Round two

Hey, how much fun was it to make the GTFS feeds for West Bus and Concord? Would you be interested in making a feed for another small New England regional system that desparately wants to publish it? A friend moved to this town and knew I might know a way to get them on the map: https://publictransportation.columbiacountyny.com/

I’m like a digital minuteman. This is my call to action. The British are coming, the British are coming! The FTA is asking for a GTFS feed, the FTA is asking for a GTFS feed! I’ve got my musket and my whisky and I’m on my way to town square to write some code for another rural transit service that feels like their website may have predated the minutemen.

This process had some improvements over the prior. First, I had two people with knowledge of the system to work with. I also had a bit more experience from the prior attempt. Nevertheless, I made another complicated python system to generate the zip file: https://github.com/columbia-county-ny-transit/gtfs-generator

One fo the people from Columbia County was able to make updates to it, which was cool. However, making a change more than moving a stop is a bit challenging. Additionally, you really have no clue if you get the change right. I have yet to find a good GTFS visualizer that is easy to install and gives you a good picture of what the heck is going on.

Despite these advantages in our process, we submitted to Google in September 2025 and only managed to publish it March 2026.

Next Steps: Absolutely laying waste to my sleep schedule

As much as I like building GTFS feeds, it gets old after a bit. Plus, I’m not going to know the exact location where the buses are supposed to stop nearly as well as the agency themselves. They should be using a tool.

Here is the list of existing tools for making GTFS feeds: https://gtfs.org/resources/producing-data/

At first, you think, “Wow, what an abundance of tools! What is bozo getting all pissy about?” Then you slowly realize that every one of these tools does something you don’t expect or hasn’t been updated in 10 years. It’s incredibly disappointing.

I bet you can see where this is going. The product of many hours of late night work has resulted in this quickly evolving tool: https://edit.gtfs.zone/

The tool is fully static web app, meaning that it stores no data on a server and is running entirely in your browser. This has many advantages. There is no login/security to worry about. It’s very reliable, because to reset it in the worst case, all you have to do is clear the storage in your browser. It also doesn’t rely on anything to be installed on your computer to run. It even sort of runs on your phone (though I have not yet optimized it for mobile). In the last week, I belive it has become useable for a rural transit agency with a standard schedule to maintain their feed with.

I would like to discuss my use of AI to build the tool. This was my first foray into developing with LLMs. Until the summer of 2025, I hadn’t really touched an LLM. I thought they were stupid and I was doing fine without them. However, I don’t think I had tried the right tools. This project was my introduction to using Claude Code. The project is perfectly positioned to be worked on by Claude Code as a static web app. I can let it stomp around in the code all over the place and as long as my manual tests succeed, then it has done its work correctly. It’s fully contained, so it can easily get a clear picture of what is going on and write code.

This does not mean that I am able to sit back and press play. I still have to be very involved in deciding what tools to use and higher level design decisions. It’s like a super speedy, but sometimes stupid developer. Context management is also very important. If you give it too much to do at once, by the end it will slow down and get all confused and start doing silly things. The great part about all of this is that it encourages me to implement many of the best programming practices. Sometimes when doing a solo project, it is easy to let yourself skip the high level planning because it’s all laid out in your head. I can’t do that here, because I have someone else reading (and writing) my code.

I should also discuss the fact that I pay $20 a month to a company that is spending lots of money to use tons of electricity and water for cooling at datacenters in places where other people live. Here is my excuse, though I am not confident in it: the inference (coming up with answers to my stupid questions) does not actually use that much electricity relative to the training (building new models). The improvement from training is becoming less and less significant. Soon, there will be less incentive to train more and just to run inference on existing models. Not the best excuse, maybe I’ll change my usage in the future.

This is all cool, but what if the bus is always running late?

No, buses are never late! Late is simply a social construct, anyways. It’s all just based on expectations. You just gotta chill out, dude.

After taking a beautiful high speed train to Florence, I had to wait an hour for my bus to the suburb of Greve. I decided to do some tourism and wander through the city to another point on the route. I arrived five minutes before the buses scheduled departure, but there was no stop to be seen. This rural bus only runs every couple hours, so I quickly installed the official app to verify the stop location. As I was doing this, I noticed a more official looking stop with a real time display on a post one block away. However, the app confirmed with Google that the stop location was on this block. I figured I should check the other one out, and I was close enough I could run back if the bus did come.

The realtime display did not show my bus at all. Huh. There was a younger person at the stop, and I hoped that they spoke English. Unfortunately they didn’t but I did some point and shoot with my phone and the bus number and they gave me a slight bit of confidence the bus would stop there. However, it was already 15 minutes late. People came and went from the stop. I was the only permanent fixture.

I felt like a fool.

I was relieved of my foolish feelings quite a bit later when a bus did show up. However, if I had just known the correct stop location and that the bus was running late, I would have just done a little more tourism, relaxed, and caught my bus when it came.

There is a second part to GTFS that I haven’t yet discussed. When you have a delay, construction, or some realtime event, you might have to deviate from your scheduled GTFS. This is where GTFS realtime comes in. If you look on Google Maps, you know that a transit agency supports GTFS realtime if an upcoming trip reads “on time/x mins delayed” instead of “scheduled”. I will discuss this more in the future, but you can demo/test the tool I am building for GTFS realtime here: https://gtfs.zone/realtime-setup/ and read more about the system behind it here: https://git.kcfam.us/gtfs.zone/deploy-gtfs-rt/

Comments

You can comment on this blog post by publicly replying to this post using a Mastodon or other ActivityPub/Fediverse account. Known non-private replies are displayed below.

Load CommentsOpen Post