Introducing KAN: An OSS project for Creation and Management of Computer Vision Edge Al Applications

To address challenges of building scalable computer vision AI solutions, we are releasing KubeAI Application Nucleus for edge (KAN - in Mandarin means “to watch”, “to see”) - a Kubernetes-native solution accelerator that enables you to easily develop, orchestrate, and operate computer vision Al applications for the edge with full control and flexibility.

Edge AI applications allow organizations to easily extract actionable insights from unstructured data streams right where the data is generated, enabling the creation of environmentally aware business solutions. Parking operators can improve parking lot utilization by analyzing vehicle patterns. Retailers can improve store operations, and customer satisfaction by continuously analyzing customer behavior in-store. However, as organizations increasingly rely on edge AI to process data closer to the source, developers and solution operators face the challenge of developing and operating scalable, distributed Al applications across heterogeneous and hybrid edge environments.

End-to-End Development and Operation Environment

KAN reduces the complexity and radically simplifies the process of building AI solutions at scale. It does so by providing a single self-hosted place for both developing AI Applications (what we call AI Skills) and then deploying and operating such applications across all your edge environments.

 

 

With KAN, you have your own no- to low-code portal experience as well as APIs that you can use to develop custom AI applications in a matter of minutes. Your custom-developed application can ingest camera and sensor data, use AI models and various other processing techniques to analyze unstructured data and then export your structured output to your desired location locally, to other environments, or the cloud, all this happening close to your data source. When building your applications with KAN, you can leverage pre-built models from our partner’s Model Zoo or create your custom ML models with Azure Custom Vision or bring your existing ML Models developed externally. Your custom-created AI application can run accelerated on x64 CPU, Nvidia dGPU, Nvidia Jetson, and Intel iGPU out of the box.

 

 

KAN is designed with machine learning operations (MLOps) in mind, providing support for active learning, continuous training, and data gathering using your ML models running at the edge. It seamlessly integrates with standard technologies such as Dapr, MQTT, ONNX, Akri, etc. As a self-managed solution, you can host it on your Kubernetes clusters anywhere across on-prem, cloud, and multi-cloud environments. It natively supports Azure Edge and Al Services like Azure loT Hub, Azure IoT Edge, Azure Cognitive Services, Azure Storage, Azure Arc, etc.

Get Started Even with no Azure Subscription or IoT Devices

Don't wait, try KAN today and experience the ease of developing and operating edge Al applications. You don’t need any edge hardware, you can get started with a few commands even without an Azure Subscription.

• Check out our GitHub Repo to learn more and Get Started Guide to start your journey.
  
• Watch a
  to learn more about our concepts and user experience.
  

Continue reading...