Orkoi Plans and Rail

Project Overview

This project started out as a ChatGPT wrapper for generating travel itineraries for any cities, back when OpenAI had just released their API and I wanted to toy around with it. Later I also built a fully functional train journey planner for all European destinations accessible by rail. This project made it to the “MVP” stage but I discontinued working on it after running into too many roadblocks.

Short pitch:

people pay a lot of money for “slow travel” scenic train journeys. A trip on vacationsbyrail.com averages 3-10k per trip!
On the other cost scale, a regular train ticket is not particularly comfortable or scenic.
What if you could pay a little more to have a comfortable ride, have a pre-planned long distance journey with several stopovers in interesting cities, get recommendations, hotels, and restaurants booked for you in a complete “tour package” for a fraction of the cost? That’s what orkoi Rail set out to be.

Challenges

It’s either impossible or prohibitively expensive to partner with travel data brokers until your site has significant traction or you’re willing to pay tens of thousands for API integration. I fetched / scraped all data from public sources
It’s quite expensive to buy SERPs (search engine scrapes). Building the scraping infrastructure yourself is a pain.
Tricky to get consistent output from LLMs that I could parse and visualize.

Technical Implementation

Serverless React app, using Vercel serverless edge functions for everything
Lots of manual data processing and cleaning, e.g. fixing mistakes in train station names. Did you know Austria has around 20 thousand train stations?
Routing and live train info fetched from an undocumented API of a major train provider
ChatGPT API (gpt-3.5-turbo)

Impact