IGPS | Devpost

Inspiration

We wanted to implement a simple and performant algorithm to visit Switzerland by train

What it does

From a list of towns, it makes the best Tour to visit them, taking into account the costs and distances and saves interest points into a database in order to create an intelligent chatbot in a next version. However, we did not have time to make the integration and the whole database so they should be considered in a upcoming version.

How I built it

With my team :-)

Challenges I ran into

SBB challenge

Accomplishments that I'm proud of

As far as I am concerned, I am proud to have learnt blockchain bases with Mr Sparks. As it was our very first Hackathon, we did not really know how to proceed, so we had difficulties to reach the final goals (we were tired). I also met people from other universities.

What I learned

I learned what blockchains are, what is the state-of-the-art in ML at the StartHack beginning ceremony. I learned PHP which I did not know so far.

What's next for IGPS

It is the section where we have the more to say. As we had few time, ressources and some bugs (API,...), we have thought about strong possibilities for the future :

With the points of interest (POI) given by the user and their durations, we could estime the time spent in each town. By knowing the trains schedules (which we didn't get so far), we could solve the TSP knowing the duration of the journey and of the visit and knowing the cost of each trainpath (making some ponderation between cost and duration). By storing the POI and labels associated with it (example: you enter StartHack as POI and university;Sant Gallen as labels), we would like to create a chatbox that would extract keywords in the user's message, "convert" them in labels and thus, associate with the POI. An other interesting perspective that we would have loved to implement if we would had been awarded (far) more time is involving neural nets as follows : during the first part of the software's "life", basic TSP's as described before would be solved and stored. Then, after a certain number of uses, TSP stored would be clustered in a small number of groups (5-10) to define -let's say 10- "standard" TSP's that represents the most frequent tours made by th customers. After that, at each use, rather than solving a TSP, a neural net would take as inputs the clients mentioned POI and "convert" them in one of the 10 "standard" tour that fits the best their expectations. So, in a marketing point of view, it could help to put efforts on these particular lines and schedules.

In conclusion, we believe, that with more time, we could have build a really great application.