The work done so far as seen from the moon.
This document describes the early design of a software platform to enable the delivery of affordable artificial intelligence (AI) services to the manufacturing industry as envisaged in the KITT4SME Description of Action. A service mesh, multi-tenant, cloud architecture is proposed that assembles AI components from a marketplace into a tailor-made service offering for a factory, connects these components to the shop floor, and enables them to store and exchange data in an interoperable, secure, privacy-preserving and scalable way.
We have been following an iterative, experimentation-driven approach to designing and documenting the platform architecture. Taking the functional decomposition in the Description of Action as a starting point, we have built a platform prototype by iteratively considering requirements from the user journey, the security and AI analysis tasks (from WP1) as well as the integration of partners' software. In addition to testing out our design ideas, this prototype has allowed us to investigate some scenarios which we deem representative of real-world use cases. Feedback from implementation tasks and regular reviews have been instrumental in informing design decisions and refining the platform design.
This work has resulted in the following contributions to the KITT4SME project:
- A set of interlocked architectural viewpoints describing the technical aspects of the platform architecture and presented in this document.
- A companion living architecture document publicly available in the KITT4SME online repository. We will update the online version of the document as we refine the architecture in response to feedback from forthcoming implementation tasks.
- A fully functional platform prototype also publicly available in the KITT4SME repository with accompanying installation instructions and software documentation.
These contributions and the present document are intended for software developers. In particular, the prototype and the architecture document aim to offer sufficient guidance for an initial platform implementation. Additionally these contributions should appeal to AI developers and open call applicants alike in that they can learn about the platform through direct experimentation with working software.