Work

I co-founded Tilt and lead engineering and product work. We build legal software for Australian and UK teams, which mostly means turning complicated work into something people can follow, check, and get through without losing the important bits.

Mobius is our work on long-running legal agents. The agents use tools and sandboxed environments when a plain language-model response isn't enough. I work on the product, the agent infrastructure, and the evaluation harnesses that check legal meaning, process, evidence, citations, and lawyer oversight together.

With Tilt, I worked on a regulatory data platform that turns AASB S2, ESRS, and related ESG frameworks into a model lawyers can maintain as requirements change. The early product reached an MVP in three months and its first paying client in six.

I converted jurisdiction, ownership, entity-classification, and exception rules into testable decision logic with legal explanations. It was a useful reminder that the difficult part is usually not the decision tree; it is figuring out which distinctions actually matter.

I built a structured Markdown corpus and MCP server over Victorian Legal Aid information, with search, page listing, content retrieval, and a proxy for public HTML. It is exploratory work on making public legal information easier for people and AI systems to use.

At Monash and in HSF's Digital Law Group, I looked at what happens when rights, obligations, prohibitions, and interpretive rules are moved into code. The research was about preserving legal meaning, not pretending that vague or contested concepts become simple because they have been formalised.

I applied that work through an executable Accord Project asset-sale agreement: a natural-language template, a Concerto typed model, and Ergo logic. It made the trade-offs much more concrete, especially when code and the parties' intended meaning point in different directions.

My Monash and LSE work gave me the technical and policy foundations I still use: statistics, deep learning, data science, cyberlaw, data protection, and the institutional effects of technology. The useful bit was learning to separate what a system can do from what it should decide.

I was invited to join the 2025 AI & Legal Practice symposium at Koiki Mabo Chambers, alongside the Victorian Legal Services Commissioner, senior counsel, and legal technologists. The conversation was about the practicalities and limits of AI in legal work, not just the headline capabilities.

Earlier at Herbert Smith Freehills, I trained lawyers on software development and project management, and wrote and researched on technology law. That experience still shapes how I think about building tools for people who need to understand, challenge, and use them.