The AI Services Framework provides a faster and safer way to take AI solutions from idea to reality.
AI activity across UNSW is accelerating. From student-facing chatbots to analytics tools and multi-model applications, teams across the University are building AI solutions that connect with UNSW data and applications.
What is an AI solution?
An AI solution isn’t just general or personal use of AI – it involves building, integrating or deploying AI capabilities within UNSW’s systems or processes.
An AI solution could be:
- a chatbot or agent connected to UNSW data or systems
- an AI tool embedded into a business process or platform
- a model that makes or supports decisions
- an AI service built for staff, students or researchers.
An AI solution isn’t:
- personal or exploratory use of approved AI tools
- the use of ChatGPT or Copilot for individual tasks such as drafting text or analysing documents.
Until recently, UNSW teams building AI solutions had to research which type of AI system would best meet their needs. Although the University has design and governance processes, they weren't built with AI front of mind. Without AI-specific patterns or guidance, projects were taking time to align AI solutions to existing architecture and review requirements, often duplicating effort along the way.
The framework
The AI Services Framework (AISF) is changing that. Developed in collaboration with Accenture, AISF gives teams a structured, reusable way to design and review AI solutions with safety, privacy and governance guidance built in from the start. The framework brings together three core components:
- design patterns that show how to structure different types of AI solutions
- a reference architecture that maps out the standard technical building blocks
- a capability model that defines a common language needed to design, deliver and govern AI solutions.
Three design patterns are available, covering single agent systems, multi-agent orchestration and third-party tool integration – with more in development and work underway to keep them up to date.
Rather than starting from a blank page, teams choose the pattern that meets their needs and build from there. For low-risk applications, such as AI recommendation systems or personalised content generation, these design patterns can cut build time from 10–12 weeks to 4–6 weeks.
Proof in practice
Project Scout, a 24/7 AI assistant designed to help students navigate enrolment, timetables and administrative processes, was the first initiative to use AISF end-to-end. Rather than creating everything from the ground up, the Scout team started from the framework's templates and tailored them to their needs.
Design and documentation effort dropped by about half, with a significant portion of the solution documentation drawn directly from prebuilt templates. The project moved through its enterprise architecture review with zero rework requests – a process that can add weeks to a timeline. Scout has since received Architecture Review Board approval and progressed into delivery.
A living framework
AISF is live and available now, but it's designed to evolve. Lessons from early adopters such as Scout are already feeding back into the next iteration of the patterns and guidance. As more projects put the framework into practice, the resources will continue to improve. Community feedback is shaping what comes next and contributions from across the University are welcome.
Whether you're proposing an AI solution or simply curious about how UNSW is approaching AI delivery, you can explore the framework on the AI Hub’s AI Services Framework page.
- Log in to post comments