MTA MetroPulse is an event-driven dispatch engine that tracks subway workers and incidents deep underground where GPS doesn't reach. It ingests simulated telemetry via Apache Kafka, triages hazards in real-time using a PyTorch computer vision model, and relies on PostGIS to route the closest qualified personnel to intercept moving trains. It's built for transit operations teams who need to transform chaotic, high-velocity data streams into split-second, geospatial dispatch decisions.
“I assumed the hardest part would be the real-time Kafka streaming, but it turned out to be the geospatial logic. Building the spatial fallback query—teaching the system to dynamically expand its PostGIS search radius when workers were too fatigued—was the "aha" moment where it's-working moment that brought the whole architecture to life.”