About
I build machine learning systems, then try to prove them wrong
I build machine learning systems, then spend most of my time trying to prove them wrong. A model that sounds confident is the easiest one to trust and the easiest one to be fooled by, so the work that matters is rarely the model itself. It is the questions around it. How do you know it is right, what happens when it is not, and who is affected if nobody checks.
That interest started in statistics and anomaly detection, where being wrong is expensive and rare events are the whole problem, and it has stayed with me through speech, language models, and production infrastructure. The through line is evaluation. I would rather ship a narrow system whose failure modes are documented than a broad one whose behaviour nobody can characterise.
Current focus
I am the founder of Biona HQ, a speech intelligence practice, and Nora Health, which turns a clinical consultation into a structured note entirely on the clinician device. Both exist because the organisations with the most to gain from speech automation are usually the ones that cannot send audio to a hosted API.
Alongside that I consult on production machine learning systems, and I coach at Codebar, which remains the most direct way I know to check whether I actually understand something.
Experience
Experience
Nora Health
2024 to presentFounder, Machine Learning Engineer
On device clinical documentation. Speech recognition, grounded note generation, and the evaluation work that decides what the system is allowed to say.
Biona HQ
2023 to presentFounder
A speech intelligence practice delivering edge native systems inside regulated environments, where the audio cannot leave the building.
TELUS Digital
Consultant
Applied AI consulting on production systems, model quality, and the gap between an evaluation result and a deployed one.
Reality AI
Machine Learning Engineer
Applied machine learning and research engineering, from experimentation through to systems running against real data.
Amazon
Data Science
Analysis and modelling at a scale where the constraint is rarely the algorithm and almost always the data pipeline underneath it.
Codebar
Coach
Teaching programming to people entering the industry. The fastest way to find the parts of your own understanding that are thinner than you thought.
Education
Education
MSc, Data Science and Artificial Intelligence
Thesis on generative adversarial approaches to fraud detection in property transactions, working on anomaly detection where the fraudulent class is rare and actively adapting.
BSc, Statistics
Statistical inference, experimental design, and multivariate analysis. The grounding that still shapes how I approach evaluation and uncertainty.
Philosophy
Philosophy
Most machine learning failures are not modelling failures. They are evaluation failures that went unnoticed until production found them. So I start from the question of how the system will be measured, and design backwards from there.
I also think restraint is underrated in this field. A system that does one thing dependably, and says clearly when it cannot, earns more trust than one that attempts everything and is occasionally wrong in ways nobody can predict.
Values
Values
Evaluation before architecture
If a project cannot state how it would know the system is wrong, that is the first thing to build. Everything downstream depends on it.
Constraints are information
On device, offline, and regulated requirements narrow the design space, which usually makes the right answer easier to find rather than harder.
Say what the system cannot do
Stated limitations are a feature. They are how a user calibrates their trust, and they are the difference between a tool and a liability.
Explain it until it is obvious
If I cannot make an idea clear to someone outside the field, I do not understand it well enough yet. Teaching is part of the engineering.
Contact
If you are working on something where being wrong matters, I would like to hear about it.
I am open to consulting engagements, research collaborations, and conversations that do not have a clear outcome yet.