ridegraph¶
Quickest routes, live closures, travel-time maps, delivery rounds, learned arrival times and ride dispatch on central London's road network, with every algorithm written from first principles and checked against a reference.
Open the live map Read the code
At a glance¶
| Map | 72,637 junctions and 127,286 road segments of central London from OpenStreetMap |
| Searches | Dijkstra, bidirectional, A*, Contraction Hierarchies, Customizable CH |
| Speed | Contraction Hierarchies settle 97 junctions per query against Dijkstra's 37,822 |
| Checked against OSRM | Travel times within 1% on 84% of 1,000 random trips, median difference 0.23% |
| Live traffic | TfL closures folded in by re-customising the hierarchy in about 20 ms in the browser |
| Learned ETA | Error on held-out trips cut from 27.5% (router alone) to 6.8% (trees) and 6.6% (graph neural network) |
| Dispatch | Hungarian matching, ML demand forecast, shared rides and surge pricing compared on the same requests |
| Quality | 92% line coverage, property-based tests, mypy, Node tests holding the browser to Python |
Where to go next¶
- Architecture: how data flows from OpenStreetMap to the browser, and what each module owns.
- Design decisions: the choices that shaped the project and what they cost.
- Results: what the numbers say, in plain language.
- Command line and Python API: the reference.