Systems architect Systems work Unified operations across 6+ services, removing over 90% of manual staff work Adaptive learning platform built around AI tutoring agents AI-assisted exam generator, awarded second place See the client work ↓ . Mathematician (M.Sc.) Academic work Turning vector bundles, Algebraic & Geometric Topology (2024) Contributor, Humanity's Last Exam M.Sc. Mathematics, University of Melbourne (2020) First-class Honours Mathematics, University of Glasgow (2016) 8 years of university mathematics tutoring, 200+ students See the academic record ↓ .
I put AI into systems that need to work.
Agents are unpredictable.
Software has to be predictable.
I build systems that use unpredictable agents without inheriting their unpredictability. Mathematics and engineering constrain, verify and compose their work into reliable software.
These are production systems solving real problems.
1.Morphisms
Premise. A state-of-the-art LLM, given a task of the right size, does it very well.
The issue comes when one tries to do large, ambiguous tasks all at once.
Result. Making the local property global is largely an engineering problem: compose small verified jobs into large verified work, with as few errors as possible, and with the human included exactly where needed.
Morphisms is that engineering. The movement and organisation of tasks are fully deterministic and schema-enforced. Stochastic output gets pushed to the leaves of the system - and at every leaf, the agent owes evidence in a required shape.
Rigid deterministic systems restrict the possible paths a piece of work can take. Real software development is not one-dimensional, or even a simple composition of loops. On the other hand, letting an agent do whatever it wants is also problematic. Morphisms is orthogonal to this trade-off. It allows near-infinite workflow possibilities for a single piece of work, inside a state-space that is mathematically precise and programmatically compiled.
The primitive: a piece of work (a record) in an initial state. The contract declares edges that can progress the work to its next state. To cross any edge, an agent must first submit evidence. This evidence must be of the correct shape (schema). The wrong kind is refused. Only when evidence of the correct shape is submitted can the agent progress the work to the next state.
This is the foundation for Morphisms - composing this primitive gives a structured and deterministic substrate that agents must move work through. Each contract is written in a new domain-specific language called Morphisms, .mor. This is compiled into a Rust CLI and database, creating a deterministic, schema-enforced system that agents must fit within.
One graph: one initial state, one terminal state. The graph branches into lanes of different rigour. Agents decide which lane a piece of work takes based on their configured instructions.
The full system includes byte-level event replay, high parallel throughput, products of graphs and built-in agents that continuously improve the workflows.
See the working demo for a more formal walkthrough. The earlier first-build log explains the initial system in more detail.
Planned open-source, currently in development. If you want early access, email me at [email protected].
2.Client work
Current work
I work with one client at a time. I am currently fully allocated.
The current work is a unified operations system for a 7-figure education business. It reconciles sales, payments, course access, email, assessment submissions and event bookings across 6+ services.
The result is one source of truth for the business. It replaces workflows that had begun to break under scale and removes over 90% of manual staff work.
This is live infrastructure for daily business operations, not an AI prototype. The work is largely reconciliation, reliability and making sure the right thing happens across every connected system.
I am currently full. Contact me at [email protected] to be added to the waitlist.
Prior work
- The Infinity Verse. An adaptive learning platform built around AI tutoring agents. View the platform.
- AI-assisted exam generator. Built during a hackathon and awarded second place. View the project.
3.Community
A community of 1,500+ builders working with claude code - orchestration, memory, production agent systems. Free, for now.
The guides are plain markdown. Read them in a browser, or curl them straight into a terminal:
curl guide.zenaitutoring.com/prd-protocol.mdJoin the community · browse the guides
I publish Morphisms development updates and other AI work on YouTube.
4.Academic work
Research and qualifications
My academic work was in algebraic topology, particularly vector bundles.
- Turning vector bundles. With Diarmuid Crowley, Csaba Nagy and Huijun Yang. Algebraic & Geometric Topology 24 (2024), 2807-2849. Article · PDF · DOI
- Humanity's Last Exam. I contributed mathematics questions to the expert-level benchmark, published in Nature in 2026. Paper and contributor list · Benchmark
- M.Sc. Mathematics, University of Melbourne, 2020.
- First-class Honours Mathematics, University of Glasgow, 2016.
Mathematical exposition. Strange Pattern in Symmetries - Bott periodicity is a video I made to explain the eightfold pattern in the symmetries of spheres without assuming a technical background.
Mathematics teaching and pedagogy
Zen MathMind Tutoring. I have tutored university mathematics since 2018 - over 8 years, more than 200 students, and roughly 3,000-5,000 hours of advanced mathematics. Students have ranged from first-year undergraduates through to PhD students.
Alongside private tutoring, I was a formal Mathematics Tutor at the University of Melbourne for five years, from 2018 to January 2023. I taught and marked Calculus 2, Linear Algebra, Engineering Mathematics, Real Analysis, Vector Calculus and Complex Analysis.
Pedagogy. No one tells you this before you learn math is a video about balancing intuition and repetition when learning mathematics.