Microsoft has announced KubeAI Application Nucleus for edge (KAN) to simplify the process of building scalable computer vision AI solutions. KAN is a Kubernetes-native solution accelerator that empowers developers and solution operators to easily create, orchestrate, and operate computer vision AI applications for the edge with full control and flexibility. In this article, we will explore the role of KAN and how IoT companies can incorporate it into their workflow.
Edge AI technologies help businesses by gleaning useful information from unstructured data streams in real time. Retailers, for instance, can improve store operations and customer happiness by continuously evaluating client behavior in-store. Likewise, parking operators might benefit from tracking car patterns to maximize parking lot occupancy. Nevertheless, as businesses increasingly rely on edge AI to process data at the edge, developers and solution operators are faced with the problem of creating and managing scalable, distributed AI applications across heterogeneous and hybrid edge settings.
KAN streamlines the procedure of creating AI solutions at a large scale by offering a unified, self-hosted platform for developing AI applications, deploying and maintaining them across all of the edge environments. Thanks to KAN, developers and solution operators can create customized AI applications in mere minutes by utilizing APIs, drawing from pre-existing models from Model Zoo or creating their own custom ML models with Azure Custom Vision, or by importing their existing ML models that were created externally.
Source: https://techcommunity.microsoft.com/t5/internet-of-things-blog/introducing-kan-an-oss-project-for-creation-and-management-of/ba-p/3725276
Close to the data source, KAN allows the custom-built application to take in camera and sensor data, analyze it with artificial intelligence models and different processing techniques, and then send the resulting structured output wherever you want it โ whether that’s on-premises, in another environment, or in the cloud. KAN was developed with MLOps in mind, meaning it can help you with active learning, continuous training, and data collection with the ML models even when they’re not in the cloud.
KAN smoothly integrates with commonly used technologies, such as Dapr, MQTT, ONNX, and Akri, among others. Being a self-managed solution, KAN can be hosted on Kubernetes clusters in various environments, including on-premises, cloud, and multi-cloud. It natively supports Azure Edge and AI services, such as Azure IoT Hub, Azure IoT Edge, Azure Cognitive Services, Azure Storage, Azure Arc, etc.
KAN can be used in the context of IoT to create and deploy computer vision AI applications on the edge. The platform’s ability to process camera and sensor data at the edge and analyze it with AI models can provide valuable insights for a range of IoT applications. For example, using KAN allows retailers to analyze customer behavior in real-time, optimize store layouts and product placement, and improve overall customer satisfaction. The same approach can be applied to a range of other IoT applications such as smart cities, transportation, and healthcare, where processing data at the edge can provide valuable insights.
In conclusion, KAN is a game-changer for developers and solution operators seeking to develop and operate scalable, distributed AI applications across heterogeneous and hybrid edge environments. By providing a unified platform for developing, deploying, and operating computer vision AI applications on the edge, KAN simplifies the process and provides full control and flexibility for developers and solution operators. With support for various processing techniques and integration with standard technologies, KAN allows organizations to extract valuable insights from unstructured data streams generated at the edge, leading to optimized operations and increased customer satisfaction.
