Work / RailGPT · 2024
Natural-language train planning for Dutch rail
Turned rigid timetable search into a conversational planning experience grounded in real Dutch rail data.
Problem
Planning train journeys often means translating a real-world question into rigid form fields, routes, transfers, and timetable constraints.
What I built
I built a conversational interface over the NS API so users could ask journey questions naturally and receive grounded, structured travel answers.
- Natural language journeys for departures, arrivals, and transfers
- Conversational UI layered over a real transport API
- Grounded answers that reduce timetable-search friction
- Shareable query URLs so travel answers can move between channels
Stack
- OpenAI tool calling for structured queries
- NS API for real-time Dutch rail data
- Next.js streaming responses
- TypeScript for end-to-end type safety
Outcome
Shipped as a working reference for how conversational product experiences can sit on top of structured travel APIs.