CyvlCyvl built an observability foundation for the cities it serves
Every year, armies of summer interns used to walk city streets with knee pads, rulers, and clipboards—crouching down to measure cracks in the pavement by hand. Reporting was inconsistent, coverage incomplete, and accuracy depended on how carefully a college student could kneel in summer heat.
Cyvl was built to fix that. Founded by engineers who studied autonomous vehicles at Worcester Polytechnic Institute in Massachusetts, the Boston-based startup mounts LiDAR sensors and 360-degree camera arrays on vehicles to drive through cities multiple times a year. With measurements accurate to within approximately two centimeters, Cyvl tells city planners exactly where to spend limited public works budgets to get the most infrastructure value per dollar.
As John Pignato, Software Engineer for Infrastructure at Cyvl, puts it: “If you can see it from the road, we can grab it.”
Cyvl has grown rapidly since John joined roughly two years ago—growth that was only possible because of a fundamental shift that put observability and Grafana Cloud at the center of its infrastructure operations. They now process data for more than 500 cities, with a 30% to 40% reduction in compute costs per job, and they have the insights they need to act immediately when something goes wrong.
“When a city council member demands an answer within a couple of weeks, we can’t afford to lose 12 hours to a data processing error,” John says.
From handoffs to pipelines
When John joined Cyvl, the data processing pipeline was functional, but also entirely manual. Each stage—preprocessing, machine learning inference, post-processing, delivery—was owned by a single person.
“Before Grafana,” John says, “we had logs—only logs—and we’d find out about something breaking the next day.” Debugging required downloading production data that was often too large to fit on a single machine. With close to 100 machine learning models running across thousands of miles of data, there was no practical way to catch anomalies before they became expensive.
Building observability for infrastructure at scale
The team had self-hosted Grafana OSS before, but they moved to Grafana Cloud because they couldn’t afford to keep spending time maintaining an observability stack. They did hit an early obstacle that gave the team pause—an obstacle that ultimately showed the true value of moving to the managed platform.
Cyvl has an unusual pipeline. Jobs burst from zero to thousands of Kubernetes nodes simultaneously, and that led to unsustainable metric cardinality. Thankfully, Adaptive Metrics provided the solution by automatically identifying and suppressing unused metric series.
Cyvl reduced their Grafana Cloud spend by 73% without spending days manually tuning configuration. “At a startup, time is your most valuable resource,” John says. “We didn’t need to spend weeks tuning cardinality. We could just shut off the metrics we don’t care about.”
With cost under control, the team instrumented their pipeline with OpenTelemetry. They also used Grafana Assistant to quickly and easily generate dashboards from natural language descriptions. This accelerated adoption across the organization, allowing engineers to build working dashboards in minutes.
Observability as an organization-wide language
Grafana Cloud isn’t just an engineering tool at Cyvl. It’s how the company operates. Product managers use dashboards to track feature engagement, while sales teams monitor platform usage. Operations tracks sensor locations and data processing progress in real time.
When something goes wrong, Grafana Cloud IRM delivers the alert into Slack with a summary and links to the relevant dashboards. From there, Cyvl’s custom Slack AI agents take over. The agents connect to Grafana through the MCP and query Grafana Cloud Logs, Grafana Cloud Traces, and Grafana Cloud Metrics
“Something goes off overnight and an agent kicks on,” John says. The result is a system that knows the difference between what needs a person’s immediate attention and what doesn’t. Most issues are resolved before anyone is woken up, and the on-call team only gets paged when a problem requires human judgment.
Scaling for the future
As Cyvl moves into more large-city engagements, the team is deepening its investment in Kubernetes infrastructure monitoring. Their burst batch processing model pushes Kubernetes to limits most organizations never encounter, making it critical to know when infrastructure is nearing its capacity.
The team also plans to extend its AI-native operations workflow. By expanding how Slack agents use the Grafana Cloud API to query logs and metrics, Cyvl aims to provide even more automated coverage, ensuring faster resolution as the volume of city projects grows.
For teams early in their observability journey, John’s advice is to make OpenTelemetry adoption as frictionless as possible.
“Once someone builds a dashboard, it’s an ‘aha’ moment—they can see into their app completely,” he says. “They understand that if they import those three lines of code, they can save themselves so much heartache.”
