No-Delay Transfer
Public Transportation Open Data Challenge 2025DiscontinuedOct 2025 — present (personal project)
Overview
"Trains are running normally right now"... Have you ever trusted those words and ended up late?
Existing transit apps only tell you about "current" delays. This app predicts future delay risk and gets people who "absolutely cannot be late" to their destination safely. It accumulates and analyses months of real-time operation data to visualise statistical risk.
Tech stack
Python 3.12
FastAPI
PostgreSQL
OpenAI API
Vite + JS
Tailwind CSS
GitHub ActionsKey features and technical highlights
1. Custom graph search engine (built from scratch)
Instead of relying on existing APIs, I implemented my own route search engine based on Dijkstra's algorithm.
- Variable transfer buffer: transfer times change dynamically, so routes are proposed according to the user's risk tolerance
- Penalty-based detour search: penalties are applied to the main route to force physically different alternative routes to be derived
2. Future delay risk prediction
Months of operation data collected from the ODPT API and other sources are accumulated and analysed.
- Statistically derives trends such as "the XX Line is 30% more likely to be delayed around 18:00 on Fridays"
- Shows a warning when future risk is high, even if service is currently normal
3. AI concierge
Integrates GPT-4o-mini to provide qualitative advice that numbers alone cannot convey (e.g. "avoiding the crowds after an event ends").
Links
Contest* The demo site is no longer available.