The mathematics ofagentic systems.
We bring academic frameworks to the complexity and security of agentic systems, and we build the harnesses that make them work. Today that means helping Intellicycle develop its agentic marketplace for recycled metals, shipping Morphika, and building research tooling for science labs.
Agentic systems are growing in capability faster than the tools we have to measure whether they are tractable, safe, or correct.
We study the structural properties of agent behaviour: branching, information flow, capability boundaries, adversarial pressure. Then we build the harness around the model (tools, permissions, memory, audit) so those measures turn into systems operators can trust. The work runs on real products and research platforms, not benchmarks.
What we study
Complexity metrics
Measures of branching, depth, and decision density in agent execution graphs, turning intuition about 'how complex' into numbers operators can compare.
Capability boundaries
Formal models of what a deployed agent can and cannot reach, so capability creep stops being a surprise discovered after an incident.
Information flow
Tracking what an agent sees, retains, and emits: the surface where leakage, contamination, and prompt injection live.
Adversarial pressure
Quantifying robustness under prompt injection, hostile inputs, and counterparties that push back, not just benchmark scores.
Tractability bounds
Knowing in advance which workflows are even decidable, and which are stochastic surfaces dressed as deterministic systems.
Operational safety
Translating structural measures into thresholds, alerts, and rollbacks that an oncall team can actually act on at 3am.
Work
Agentic systems
Intellicycle
An agentic marketplace for recycled metals. We help build the agent harness and agentic features behind it.
Morphika
Our agentic email product. Every action is gated, signed, and written to an append-only ledger.
Fixr
A collaboration with the Fixr team on product design and agentic workflows for property and asset operations.
Trixode OS
Our own design system: one set of tokens and templates behind every deck, document, and interface we ship.
Scientific computing
gemicro
Agent skills and a CLI for an electron microscopy research group at UVic. Every session, simulation, and HPC job lands in a hash-chained ledger anchored to Morphika.
ElectroSim
The group's shared GPU computing hub. gemicro is the agent layer on top, so simulation and analysis run with a full audit trail.
Physics simulations
Real-time fusion confinement: ion dynamics in a Farnsworth fusor and charged particles in a tokamak, compiled to WebAssembly to run in a browser tab.
Team
A small group of science people working on interesting problems in business.
Hussien Ballouk
Physics, strategy, and software engineering, with a math backbone.
- B.Sc. & M.Sc., Southern Federal University (Russia)
- Ph.D. Physics, University of Victoria (in progress)
- Background in nanotechnology
Working on something agentic?
Tell us the shape of the workflow. We'll come back with what's measurable and what isn't yet. No decks, no boilerplate.