Diode

FlowCast turns the traffic cameras a town already has into a measured picture of its road network. It also forecasts what a road change would do before it happens. Congestion is a global problem that keeps growing. Drivers in the most congested cities lose over 100 hours a year in traffic, and cities and vehicle fleets keep expanding. Yet most towns still rely on people watching CCTV walls, occasional manual counts and signal plans retimed every few years. None of that scales to thousands of roads, or answers questions like "what happens on Thursday at 8:30 if we close this road?" Observe. The town's existing camera systems push recordings to FlowCast; no new hardware is needed. Meta's SAM 3.1 reads the video with text prompts for car, motorcycle, bus and truck, with no training per camera or city. Each vehicle is placed on the real road through the camera's calibration, and pairs of frames give its speed. Every 5 minutes, each camera reports vehicle counts, flow, speed, occupancy and congestion level for the road it watches. Understand. The observations form the state of the whole network, mapped onto a road graph built from OpenStreetMap. FlowCast tracks network figures: vehicles in town, share congested, speed index and delay in vehicle-hours. It learns each road's usual behaviour by time of day and kind of day. Forecast. A graph model fitted to the observed days predicts any future date and time: expected traffic, hotspots and recommended actions. For a what-if, such as a closure, roadworks or an event, it spreads the change through the network, reroutes traffic and reports the extra delay at each affected junction. Each case gets a verdict (Recommended / With caution / Not recommended) and action items. Ask in plain words. Muse Spark, through the Meta Model API, is the Traffic Assistant. An officer asks "close High Street eastbound tomorrow at 9am for an hour?". Muse Spark turns it into a structured plan for the graph model, then writes up the result. The numbers always come from the model, and every number in Muse Spark's answer is checked against it. Muse Glimmer runs the same assistant locally for low-cost development. Everything appears in a dashboard for traffic teams: the city overview, the forecast and what-if cases, and an onboarding page that brings a new town in from a map and its cameras. FlowCast is a working proof of concept, tested end to end on three simulated sample towns. Planned next steps are forecast-based signal timing, incident response and infrastructure planning.