BridgeAI MLOps Knowledge Hub

About this project

BridgeAI is an InnovateUK-funded, national level AI (Artificial Intelligence) acceleration programme designed to stimulate the adoption of responsible AI in priority sectors such as Construction, Transport, Creative and Agriculture, by bridging the gap between innovation and implementation.

As one of the four main strategic partners of BridgeAI, Digital Catapult plays a crucial role in the business and technology co-creation aspects of the programme. As part of the BridgeAI contributions, Digital Catapult (DC) focuses on the topic of Machine Learning in production / Operationalizing ML.

Organisations often face numerous organisational, technical and operational challenges when transitioning their ML (Machine learning) models from development to production. These challenges include complexity in integration, lack of required skills and expertise, absence of mature tools and robust frameworks for ML Operations (MLOps) and more. In response to these challenges, Digital Catapult has developed its Applied AI suite offering, which includes the following:

• A comprehensive, web accessible MLOps maturity assessment.
• An end-to-end pre-built MLOps pipeline, created using open-source tools.
  • This pipeline has been designed to give SMEs the opportunity to engage with practical tools and resources that can notably enhance their automated AI/ML offerings, which is expanded on in the MLOps Clinic.
  • It has also been designed to give SMEs the opportunity to speed up the development of their AI/ML offerings that they seek to automate and deploy, with minimal coding on their behalf.
• A corresponding ‘MLOps knowledge Hub’ that contains relevant information on practical MLOps and how to successfully build an MLOps pipeline from scratch for commercial projects, based on the creation of the pre-built pipeline.
• An in-person MLOps clinic that offers hands-on help to participants via workshops as well as interactive support from a range of MLOps providers in the market.

Why this knowledge hub?

This knowledge hub has been designed to explain the design and implementation of a viable, end-to-end MLOps pipeline for small and larger organisations. It is important to note what is within and beyond the scope of this hub:

In scope

1. The research our team had undergone and design decisions made for each component of the MLOps pipeline with explanations on why they were made, with the components comprising:
   • Data versioning
   • Model training pipeline
   • Model serving
   • Prediction service
   • Model monitoring
   • GitOps
     
     
2. Considerations for SMEs from a corporate perspective, including:
   • A Deployment Service Life Cycle framework to wholly assess considerations that such organisations should account for before, during and after the implementation of your automated AI/ML offering
   • An overview of the relevant skills and tools required to implement an MLOps pipeline, as well as a tool horizon scan to give SMEs a range of alternatives to the tools the DC team chose
   • A maturity assessment to evaluate the viability of the AI/ML offering(s)
   • An overview of the principles and core considerations surrounding MLOps, and an overview of the AWS Well-Architected Framework for further consideration on top of the principles and core concerns

In future, this knowledge hub may also cover further troubleshooting the team has undergone to implement components of the pipeline, and an expanded horizon scan should the team have to consider other tools for the components.

Out of scope

This knowledge hub does not provide any extensive information on what MLOps is, its best practices, or on the functions of individual components of an MLOps pipeline. It does, however, provide links to platforms that address these points.

The links to these platforms can be found in the Horizon Scan page, and the Best Practices page of this hub.

Scalability, Security and Compliance of the MLOps pipeline are also not addressed in this knowledge hub. The team plans to take care of these aspects as a part of Digital Catapult’s Year-3 offerings.