KModels: Unlocking AI for Business Applications

TL;DR

KModels tackles the challenge of deploying AI in business environments by providing a template-driven lifecycle framework that runs on-premises (and in the cloud) using Kubeflow Pipelines and KServe. It enables model vendors to package AI capabilities as immutable Templates, which clients instantiate, train on local data, and run without requiring data science skills. The framework introduces Model Templates, a Model Store, and pluggable Data Connectors/Monitors to support end-to-end lifecycle management, including data acquisition, training, deployment, monitoring, and retraining. In a real-world pilot, three models deployed within a client’s asset management system improved Failure Code suggestion accuracy from 46% to 83%, demonstrating practicality, locality, and rapid deployment potential for enterprise AI adoption.

Abstract

As artificial intelligence (AI) continues to rapidly advance, there is a growing demand to integrate AI capabilities into existing business applications. However, a significant gap exists between the rapid progress in AI and how slowly AI is being embedded into business environments. Deploying well-performing lab models into production settings, especially in on-premise environments, often entails specialized expertise and imposes a heavy burden of model management, creating significant barriers to implementing AI models in real-world applications. KModels leverages proven libraries and platforms (Kubeflow Pipelines, KServe) to streamline AI adoption by supporting both AI developers and consumers. It allows model developers to focus solely on model development and share models as transportable units (Templates), abstracting away complex production deployment concerns. KModels enables AI consumers to eliminate the need for a dedicated data scientist, as the templates encapsulate most data science considerations while providing business-oriented control. This paper presents the architecture of KModels and the key decisions that shape it. We outline KModels' main components as well as its interfaces. Furthermore, we explain how KModels is highly suited for on-premise deployment but can also be used in cloud environments. The efficacy of KModels is demonstrated through the successful deployment of three AI models within an existing Work Order Management system. These models operate in a client's data center and are trained on local data, without data scientist intervention. One model improved the accuracy of Failure Code specification for work orders from 46% to 83%, showcasing the substantial benefit of accessible and localized AI solutions.

Figures (9)

• Figure 1: The entire life span of an AI model as envisioned by KModels. It starts with the development of the model at the lab, continues with its packaging and publishing as a template. Both these steps are done on the model vendor's end and require skilled personnel . Later, on the client's end the model is instantiated, trained and deployed, requiring only business configuration and no data science skills.
• Figure 2: Architecture diagram depicting KModels as the control layer on top of two different options of stacks: Kubernetes based or OpenShift based. The key components are KServe, and Kubeflow/Data-Science pipelines. The underlying infrastructure layers assure the robustness of KModels while the complexity of managing and configuring those layers is largely hidden by KModels' control layer, reducing the expertise level required from the target users.
• Figure 3: The template's metadata (left) defines the information needed for instantiating a model and the supported arguments for configuration. Upon instantiation, a configuration file fills in all the necessary data and an model instance is created.
• Figure 4: REST interface for KModels allowing to create models, delete models, infer, provide feedback, etc.
• Figure 5: KModels architecture