The team needed a better way to detect defects in utility infrastructure using machine learning. Their previous approach relied on manual labeling workflows and open-source tools, which lacked the configuration options and support required to meet internal SLAs. Progress was slow, and scaling was difficult.
They adopted Databrewery, connecting raw data through a simple API to centralize their labeling operations. With built-in collaboration tools, they were able to quickly onboard internal and external labelers, streamline training, and increase annotation throughput all within a single environment.
By removing bottlenecks in their pipeline, the team accelerated model training by over 10x. Engineering resources were preserved for core development work, and labeling costs were cut by 50%. Most importantly, they maintained high data quality while moving models into production faster than ever.
This AI-driven company builds advanced drone technology to support safer and more efficient energy infrastructure. By equipping drones with aerial sensors and computer vision models, they enable utility companies to detect issues like vegetation overgrowth, damaged insulators, and other safety risks long before they become real hazards.
One of their core use cases is identifying dangerous electrical setups by analyzing massive amounts of inspection imagery. To train the models that power these systems, the team needs thousands of accurately labeled images. Their previous setup involved heavily manual workflows and open-source tools that didn’t offer the flexibility or performance they needed.
They chose Databrewery to centralize and accelerate their labeling pipeline. With the ability to handle tiled imagery and other complex data formats, Databrewery helped them streamline data organization and connect raw datasets using a simple API. Collaboration tools allowed internal teams and external experts to work side by side in a single environment, speeding up onboarding and boosting labeling throughput.
To push efficiency even further, the team adopted model-assisted labeling. By integrating their own model predictions directly into Databrewery, they could focus human review on edge cases and reduce the time spent on routine tasks
“We used to spend weeks just getting the labeling environment up and running,” said Lina Aalto, Senior ML Engineer. “Now we’re focused on refining models instead of wrangling tools. Once we built our first loop with model-generated labels, we cut our manual load in half.”
The time savings were significant. “Our team can now process more components and defect types in less time without increasing cost,” added Henrik Weiss, Computer Vision Lead. “It’s transformed how we scale model training. The more we iterate, the faster we improve.”
With a continuous loop in place combining model inference, human validation, and automatic feedback, the company is advancing their inspection systems faster than ever. By using Databrewery to manage the full lifecycle of training data, they’re pushing the boundaries of what computer vision can achieve in infrastructure safety and energy resilience.