The Queensland Government Customer and Digital Group run regular supplier information sessions on transforming and enabling technologies as part of their industry engagement strategy. In February 2024, WorkingMouse was invited to share our insights on ‘Composite AI & Models for Modernising Government Services’. Following the session, we’ve summarised composite AI, discussed control and given you our 5 top tools and tips in this article.

The slides from this are available for direct download here.

Composite AI

As AI is a hot topic, we wanted to discuss and show how we’re applying composite AI to government projects with our philosophy. But firstly, lets discuss what exactly composite AI is. AI is a very umbrella term, and there are a lot of different parts to AI above those large language models (LLM’s). Some of the areas are search and optimisation, deep learning, neural networks, intelligent agents, you name it. Composite AI is about recognising this range of AI tools and combining them to balance out their different strengths and weaknesses to achieve your business goal.

Hit the edit button

To achieve our business goals using composite AI, we implement pipelines as repeatable processes, ensuring consistent application of composite AI across various stages to produce reliable outcomes, (we will go into more depth about this later). Anything that has machine learning will underpin an error rate. This is just the way the algorithm works, which means there will be some level of inaccuracy. We’ve found that it’s important to hit the edit button on any of these things, because you can’t trust that the machine learning and GPT is going to produce what you want. GPT and other LLM’s may give you a first review, it might give you framework and help with analysis, but it’s not going to give us the final polished product at this point. This means we need to balance the use of AI with the use of humans. Think of it as broadening the scope of AI beyond just machine learning algorithms. Consider how these AI tools can be effectively utilised to address challenges across various workflows/pipelines.

5 Tools and Tips

How do we use composite AI for modernisation projects? We empower software teams augmented with AI and the best tools. Our five key modernisation tools can be seen below and tips for using them.

1. Coding with CoPilot

• Hit the edit button & ensure accuracy > don’t rely on AI as it has an error rate.
• Integrate AI at the lowest level of abstraction Integrated Development Environment IDE (coding level)
• Use Gen AI on what is already well defined (e.g. documentation)

A popular AI tool around coding currently is Co-Pilot. From our experience with Co-Pilot, we’ve found that whilst it does produce code, it doesn’t always compile, but it also always needs to go through the usual review processes. It offers an auto-complete feature that accelerates developers progress, but ultimately, they are still responsible for completing the task at hand. For modernisation projects, it’s about using that AI and the best tools for the team. We tested this at a higher level of abstraction and it doesn’t work, you may get some initial velocity but maintain the Codebase with not awareness and documentation is high risk.

2. Optimisation Problems

• Empower direct problem solving.

Optimisation problems are common in modernising legacy systems, we have found using AI is particularly effective in addressing these problems because the AI tool can analyse larger volumes of data, identify patterns, and suggest solutions that improve system performance. A good example of this is the hill climbing algorithm.

3. Platform Engineering & Golden Paths

• Use pipelines and automation for throughput of work and validate quality.

Moreover, we’ve found that there’s another set of technologies which you may not have heard about yet, which is starting to get some serious airtime because it gives a lot of power for teams to produce high-quality software at scale. This area is called platform engineering. The concept of GoldenPaths ensures a consistent pathway to produce the same result each time. For WorkingMouse, platform engineering involves using models to illustrate golden paths, which are then used within modernisation projects, resulting in a seamless integration process.

4. Team Topologies

• DevOps: pipelines are the enabler that breaks down walls between development & operations.

• To build momentum build a platform team to empower your development teams.

If you’re not familiar with Team Topologies, it's worth exploring. This concept emphasises fostering effective collaboration and communication within teams, ensuring a smooth working environment. These facilities a shared understanding across WorkingMouse, enhancing the utilisation of technologies. Check out his with works with Platfrom Engineering.

5. Shared Understanding using Models.

• Raise knowledge to be a first-class artefact (this is just as important as code).
• Model-Driven Engineering (MDE) models are the enabler that build bridges across the organisation and system.

Moreover, we view everything as a model. Whether it’s a user interface, database, or business process, modelling allows us to create a common understanding. By integrating these models into the software engineering process, we can effectively scale our modernisation projects.

Supplementary

Excluded from this article but also in the presentation:

Qld Health Project Demonstration

• A like for like replacement of ieMR Configuration Portal
• Meta Models + User Interface & Entity Models
• Pipelines
• Editing locally with Composite AI
• Environments with CI/CD

Full Case Studies can be found here.

Who is WorkingMouse?

• 45 team members across 6 cross functional software teams.
• First nations Scholarship with UQ
• ISO 27001 Certified
• QGov ICT 13.3B Panel Member
• Jidoka: WorkingMouses Process for Auotmation with a Human Touch.

Please see our Government Solutions Page for more information.

Summary and Acknowledgement

Implementing these tools has proven beneficial as they enable us to maximise the potential of AI technologies whilst leveraging human skills. However, it’s important to remember that human oversight is crucial to ensure the accuracy, functionality, and quality of a project, emphasising the importance of maintaining a balance between automation and human expertise. We would like to extend our gratitude to Matthew Rose, Manager of Industry Engagement at Department of Communities Housing and Digital Economy for facilitating the supplier information session.