eSmart Systems has launched AI Studio, a software platform that lets utilities and technology firms create and deploy custom computer vision models for grid inspection through a web interface and application programming interfaces.
The Norwegian company, known for visual analysis of electricity infrastructure, is making its computer vision tools available as modular services for the first time. AI Studio is positioned as a full-stack development and deployment environment that can sit alongside existing inspection and asset management systems.
Computer vision has become a common feature in inspection programmes that use drones, helicopters and ground crews to capture images of transmission lines, distribution equipment and substations. However, operational requirements vary by region and network owner. Defect definitions and inspection standards also change over time, making it difficult for off-the-shelf models to work consistently across large fleets of assets.
AI Studio centres on eSmart Systems' patent-pending Adaptive AI methodology, which uses few-shot learning so detectors and classifiers can be built from a small set of images. Models can also adjust to new definitions and edge cases without traditional retraining cycles.
eSmart Systems reported benchmark results of around 99% accuracy and said the approach can deliver up to 35 percentage points higher accuracy than competing few-shot methods. This is validated against 6 published few-shot classification methods from NeurIPS, ICCV, and CVPR.
Custom models
Utilities have often needed specialist machine learning teams to tailor computer vision systems to their networks. That work can include building annotation processes, setting up training runs and maintaining models as standards shift or new asset types enter service.
AI Studio targets teams that want to adapt models to their own asset definitions and workflows without building an internal machine learning function. The platform is offered with both a browser-based interface and APIs for integration into other software.
Henrik Bache, chief executive officer of eSmart Systems, described the move as a shift from product delivery to a platform approach.
"We've spent over a decade building infrastructure-grade AI for utilities. AI Studio opens that intelligence as a platform," said Bache.
Three components
AI Studio groups its main functions into three areas: Model Builder, Model Garden and Pipeline Builder. Each addresses a different stage of building and using computer vision in inspection work.
Model Builder is intended for rapid creation of custom models from a limited number of example images. The workflow does not require annotation pipelines or training runs, and models can adapt as definitions evolve and new edge cases emerge.
Model Garden provides access to a library of detectors, classifiers and pre-built pipelines. The library draws on millions of images from transmission, distribution and substation contexts, refined through deployments with more than 75 utilities.
Pipeline Builder focuses on combining multiple steps into a workflow. eSmart Systems described a visual editor that can chain tasks such as detection, classification and measurement, and said the system provides input and output transparency for auditing and compliance.
Each model and pipeline receives a dedicated API endpoint, supporting integration with inspection platforms, engineering systems and drone data workflows. This also makes the tools accessible to AI agents that can call APIs as part of automated processes.
Agent-based workflows
Software suppliers across industrial sectors are adapting products for agent-based automation, where systems assemble workflows dynamically rather than relying on fixed user interfaces. eSmart Systems described AI Studio as "agent first" and API-based, linking the approach to its view that enterprise software will increasingly rely on agent-driven processes.
Erik Åsberg, chief technology officer of eSmart Systems, said the key technical difference lies in how quickly models can incorporate new defect definitions.
"What makes Adaptive AI genuinely different is that it breaks the dependency between model performance and retraining cycles. When a utility's definition of a defect changes - and it always does - the model updates in minutes, not weeks," said Åsberg. "That changes the economics of deploying AI in the field entirely."
Data and differentiation
eSmart Systems said its market position rests on long-running access to utility imagery and asset data, along with the Adaptive AI approach and a platform architecture that can incorporate improvements from newer AI models. It framed these as differentiators from general-purpose computer vision suppliers and foundation model vendors.
The company is already known for Grid Vision, a product used to inspect electricity infrastructure. AI Studio extends that work into a platform that third parties can use to build inspection workflows and integrate results into operational systems.
eSmart Systems said AI Studio is publicly available and intended for utility teams, technology companies and inspection professionals working across transmission, distribution and substation inspection.