My research program is organised around building complex, distributed software systems that can be trusted to do what they claim. Its foundation is the automation of software interactions — decomposing a process into individual, repeatable, composable tasks, then linking them into something larger. This is the technology that enables the "highly-linked" modern software systems connecting everyone and everything on the Internet today. Under this theme I have worked on communication mechanisms between software systems, distributed platform design, and the application of these techniques to non-traditional domains such as automated document processing.

What has changed is the scale of what we are now being asked to trust. Automation is no longer only about data and access to private information; it increasingly involves systems whose internal behaviour nobody can fully inspect, large language models, autonomous agents, and financial infrastructure that settles real value on distributed ledgers. The question I keep returning to is a software engineering one: how do we check? Software engineering has spent decades building precisely these disciplines, testing, verification, provenance and compliance, and my current work is about transferring them to the systems society is being asked to rely on. In particular, I run three connected programs: (i) software engineering for trustworthy AI, where the black-box and ad-hoc nature of current AI solutions can benefit from the well-established systematic frameworks of software engineering, testing and repairing model behaviour, watermarking and attributing generated content, and evaluating machine reasoning in ways that can be independently checked; (ii) trust and accountability in decentralised systems, where blockchain technology brings a new paradigm for how society can function when there is no trust, now extending to whether tokenised financial systems can be shown to settle finally and to keep functioning as markets; and (iii) machine-executable compliance, translating regulatory and legal text into constraints a system can actually check, so that compliance is verifiable rather than asserted.

My current projects involve the following topic areas:

• Testing, verification and assurance for AI systems
• Trustworthy and accountable large language models and agents
• Provenance and watermarking for AI-generated content and code
• Blockchain-based systems for enhanced privacy and trust management
• Verification of settlement and payment systems in tokenised finance
• Machine-executable regulatory and legal compliance
• Privacy-preserving, decentralised data analytics and management

At UNSW I teach data management and web applications, software engineering projects, data structures and algorithms, and blockchain application architectures.  I currently supervise 17 doctoral and master's candidates, as lead supervisor for most of them. What I aim to bring to a supervision is the software engineering side of the problem: framing it so that it can be evaluated, designing experiments that will still convince a sceptical reader, and building the reusable artifact, a benchmark, a tool, a dataset, that outlives the paper. I am particularly keen to hear from prospective candidates interested in various verification, assurance and testing issues in building AI systems.