DwellTransit
Live now at Wolfline

Real-time arrivals university riders can actually trust.

Campus bus feeds and printed schedules are wrong all the time. Dwell rebuilds the predictions from raw GPS — and proves it. One platform, built end-to-end: GPS capture → GTFS feeds → native iOS & Android apps → a self-correcting accuracy loop.

85%
prediction accuracy
better than the schedule
17,930
predictions scored
7
routes live at Wolfline

See the full accuracy breakdown →

The whole lifecycle, end-to-end

Most transit systems stitch together a different vendor for each of these stages. Dwell owns every one — and the last stage feeds back into the first.

  1. 01
    Capture raw GPS

    Buses report raw GPS — the only input I trust. Vendor 'predictions' and schedules copied in as fake live data get thrown out at the door.

    Raw GPS ingest Vendor-agnostic
  2. 02
    Generate GTFS-RT feeds

    Each position is snapped to its real route shape and published as standards-compliant VehiclePositions + TripUpdates feeds.

    GTFS-RT VehiclePositions TripUpdates
  3. 03
    Power native rider apps

    The same engine drives white-label iOS and Android apps — the arrivals students actually open on the way to class.

    iOS Android White-label
  4. 04
    Capture history

    Every observation is stored. Per-segment and per-stop timing is learned from millions of past trips, by time of day and day of week.

    Time-series store Per-segment timing
  5. 05
    Measure accuracy

    Predictions are scored against what actually happened — continuously — and published on a public accuracy page that updates daily.

    Daily scoring Public report
  6. 06
    Refine the GTFS static feed

    Accuracy gaps reveal where the published schedule and shapes are wrong. I fix the static feed — and split service into summer vs. semester calendars.

    GTFS static Summer vs. semester
  7. …and the loop closes

    Cleaner static feeds and summer-vs-semester models flow back into the history layer (04), which sharpens the live predictions in 01–03. The system gets more accurate the longer it runs.

Why summer vs. semester matters

A campus in July moves nothing like a campus in October.

Most systems average every day of the year together — and get both seasons wrong. Dwell learns separate timing models per service calendar, so a quiet summer route and a packed semester route each get predictions tuned to how that campus actually moves. It's the clearest payoff of owning the whole loop: the schedule data and the historical data improve each other.

What universities get

Standards-based GTFS-RT

Drop-in VehiclePositions + TripUpdates feeds that work with the transit apps riders already use.

White-label rider apps

Your brand and colors — native iOS and Android apps powered by the same engine.

Provably accurate

A public accuracy page, updated hourly, so riders and administrators can see the truth.

Fast & reliable

Served from the edge via Cloudflare — cached, resilient, and inexpensive to run.

Bring trustworthy arrivals to your campus

See exactly how accurate Dwell is for your system, then get a feed your riders' apps can use.