▲ 4 points

Show HN: Ask questions about the Swiss train timetable

fragplan.ch

by bekoeppel·16h ago·5 comments·view on hn ↗

I use the SBB website and app every day to find my Swiss train journeys, but even with the touch tiles I find it cumbersome to use, especially for extra filters like avoiding bus or finding trains with space for bicycles.

So I built https://fragplan.ch/ to describe a train journey in natural language, and translate it into an SBB timetable request. I'm especially happy about the many settings one can make on SBB.

The name is a play on the words "fragen" (to ask) and "Fahrplan" (timetable).

I wish there was something like this directly in the SBB app...

5 comments

I tested the use case I could've really used this for, but it doesn't seem to be able to do this: "Biel to Zürich, but not through Grenchen".

The SBB timetable app and website can become really unhelpful if there has been an "accident involving a person" [0]; other than that, I've got little issues with them day to day.

But it seems your model doesn't have a way to know that the IC from Biel to Olten goes through Grenchen (it only presents the obvious routes that wouldn't work in this case).

[0]: https://company.sbb.ch/en/railway-development/services-rus/r...

Oh I hadn't thought about that yet! But when there is a disruption like a accident or construction site, don't you get a re-route?

If it's something small, usually yes. But very often, that's still on the same routes (often, not all rails are blocked, so only some connections through the station suffer).

But I've also seen my fair share of "trains in this direction won't recover for hours" over the years, and directly after such an incident, the official systems usually don't really adapt all that well. I don't know if they're "massaged" manually after that or if it's just a case of the entire system finally figuring out that an entire station is FUBARed, not only one or two scheduled connections…

Nice. I tested a very simple example and got exactly what I hoped. Is there a LLM behind it or some more traditional kind of AI?

I built a combination of rules (e.g. HB is translated to Zurich HB) and LLM (Haiku). The rules were necessary because if I send a user to SBB.ch with destination "HB", SBB translates that to Herzogenbuchsee... not what I had expected.