FlowTransit
FlowTransit provides AI-powered route planning and optimization software for public transit agencies, private operators, and consultancies. The platform models networks from GTFS feeds and ridership data, then uses deep reinforcement learning over graph neural networks to generate and score millions of alternative route configurations in minutes. Planners can edit routes on real street geometry, compare scenarios against current service, and export results as GTFS with Title VI equity tables.
- Mobility & Transportation
- Artificial Intelligence
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Public transit networks are often designed around outdated assumptions and rarely keep pace with shifting population patterns, new developments, or changing travel demand. Manual route planning is slow, limits the number of alternatives explored, and typically requires separate travel-model runs to score each option, leaving agencies unable to optimize service within their budget.
Solution
Flow is route planning and optimization software that models a transit network from GTFS feeds and ridership data, then uses a deep reinforcement learning model over graph neural networks to generate and evaluate millions of alternative route and frequency configurations in minutes. The platform scores every candidate against the user's stated weights and constraints, such as fleet size, coverage, and travel time, returning the top three networks for review. Planners can edit routes directly on real street geometry, compare scenarios against current service, and see the passenger-load impact of each change in real time. The same engine also handles continuous planning, absorbing road closures or new points of interest and flagging patterns like bus bunching with suggested fixes, with everything exporting to standard planning tools.
Target Audience
Primary customers are public transit agencies, private operators, and consultancies that plan or bid on bus networks, from single corridor studies to full network redesigns.
Features
- Deep reinforcement learning on graph neural networks that optimizes routes and frequencies together, handling full network complexity without simplifying away inputs
- Drag-and-drop route design snapping to real streets, with headways set per line and period, and running times derived from AVL history
- Scenario generation producing millions of candidate networks in minutes, scored on speed, reliability, coverage, and cost
- Automated per-route comparison between a candidate network and the current master plan, attributing ridership shifts and service changes
- Passenger load tracking by stop and route section against vehicle capacity to flag crowding before schedules are written
- Continuous planning features that detect bus bunching or demand shifts in live data and suggest interventions, plus forecasts learned from the past year of service
- Exports as GTFS, board maps, service equity tables for Title VI analysis, and reports with maps, scorecards, and accessibility analysis