Train Journey Planner

Travel Europe By Train. See the sights, save carbon emissions, and have a fun adventure. A project by the ToTravelTo.com team.

Comment

Inspiration

European train travel is a lot of fun, but not without it's challenges. Being able to chat to AI is like having your own personal travel agent. But can we get AI to make a useful travel itinerary by train, and make sure we have fun too?

What can we learn to make totravelto.com even better.

What it does

We help the train traveller have a fun, greener, adventure.

The traveller can input their origin and destination as well as start date and the number of days they want to travel.

They will then receive an multi-day itinerary, covering where they should stay the night, and what highlights they should explore. They'll also know how long it is between each of the locations, and an overall distance travelled.

The distance travelled can then be used to illustrate the amount of carbon emissions saved, when compared with flying.

How we built it

First we decided on the inputs we needed, then we gave the date and duration some sensible defaults, to make it even easier for people to use.

Additionally, we refined the prompt to help generally define the app, and what it should do, as well as how it should use the input parameters.

There was some iteration with defining the prompt so that we didn't have too much in it, and whilst we preferred to have the pleasantries such as 'please' in, taking them out reduced the length.

It was also useful to make the prompt more specialise in some circumstances i.e. telling it to emphasise what was special about a place to visit at a location.

Finally, we found that for further information and refinement we could give people an in-depth link to our own application, so that they can get real-time schedules and prices for their adventure i.e. https://totravelto.com/copenhagen

Challenges we ran into

Some LLMs gave more variable results, but the ease of which you could try different Generative AI LLMs meant you could experiment quicker with the Party Rock platform, and it's easy to use user interface(UI). We settled on the LLama 2 Chat 70b LLM, but were impressed at the range of LLMs made available to us beyond chatGPT which we have often played around with at totravelto.com

Accomplishments that we're proud of

It was great to see that the distance maths worked with the Llama 2 Chat 70b LLM, and also could be used for further calculations to see how much carbon emissions are saved by not flying the total distance i.e. whilst there's many methods the EEA says train are 19g CO2 per km, and air travel is 123g CO2 per km, and we can use this to calculate the difference saved by taking the train.

What we learned

We saw that PartyRock is a fun way to trial out ideas with a range of LLMs quickly.

What's next for Train Journey Planner

It would be nice to expand the link back to include the full travel query, so that we can provide a full travel cost for the user of PartyRock, though for now it is great to be able to continue the conversation by linking to To Travel To via destination links such as: https://totravelto.com/copenhagen ToTravelTo.com is building a self-contained AI with planner and purchasing of train tickets, but we'd love to integrate with other systems such as PartyRock.

Built With

  • partyrock
  • promptengineering
Share this project:

Updates